Home

D10.2

1. e un ent C e un e un eut e e un Dataset type Parallel A 1 file of 100MB A 100MB A 1 A 1 file of 100MB A 100MB A 4 A 1 file of 100MB A 100MB A 8 A 1 file of 100MB A 100MB A 16 BO B B B 16 Tests Description Ne Ne we SN SN N SN UJ e Once the iRODS servers 3 2 were deployed a remote zone was created on each 1RODS servers to allow transfer of data People involved in this work were granted access to the remote zone These zones were created on the highest performance filesystem available on each server cf Tests Environment above The iRODS servers used the GSI certificate based authentication method for the users The two datasets used for the benchmark were created for the NFTT sub task by a program using random number generator to build it contain to prevent from any compressing process during the transfer steps The 1RODS servers were tuned regarding the network performances to ensure better performance in the transfer process This has been done by adapting the following parameters in the server configuration file irods server config reConfigs core re acSetNumThreads msiSetNumThreads sizePerThrInMb maxNumThr windowSize e sizePerThrlInMb The number of threads is computed using numThreads fileSizeInMb sizePerThrInMb 1 where sizePerThrInMb is an integer value in MBytes It also accepts the word default which sets sizePerT
2. This survey has 16 questions and is divided into five parts while the second to fourth all deal with data workflow aspects Data Characteristics Data Movement Data Sharing Data Post Processing Other All questions are related to resources and services which are or should be available in relation with calculations on the PRACE infrastructure The survey does not discriminate between possible differences for Tier 0 and Tier 1 systems In case differences are expected please indicate that To better understand those needs on data we ask questions about the specification of the data and the workflow This workflow covers in more detail questions like Where does your data come from Data Movement What sort of post processing are you performing Post Processing and Data Analysis Where do you store your processed data Data Movement Should your processed data be made available and to whom Data Sharing Could you describe the storage size and type used for each phase Data Characteristics e Would you like to enhance your workflow and how could this be achieved Other Final Remark If any of the questions sounds too technical to you just note that and do not answer Data Characteristics Nature of the data 1 1 How can your data be described number of files average size of a file meta data information type of data ASCII or Binary optionally more details for binary data images sound files 1 2 How woul
3. 78 MB s R W iRODS 3 2 Cloud Virtual Machine NIIF e 2 cores e 2GB e 100 Mb Ethernet card e Debian GNU linux 6 0 Local FS XFS 13 MB s R W iRODS 3 2 Methodology The methodology used in this evaluation intends to use the methodology defined in the New File Transfer Technology task However this methodology was not fully applicable due to the reduced capacity testbed that was provided by some partners The initial methodology was defining the following set of information e Similar operational conditions minimum requirements o TCP buffer sizing o MTU and Jumbo Frame o Disk performance no bottleneck o Network Capacity using PRACE dedicated 10 Gbe network e Specific Test Case o Dataset A small files B large files o Workload A 100G B 500G C 1TB o Parallel Streams A 1 B 4 C 8 D 16 e Performance reference value with gridFTP in the same configuration Each test must be run several times to deliver a reliable measurement Test Case Run Dataset type Parallel streams 1 A 1000 files of 100MB A 100GB A 4 D A 100 files of 1GB A 100GB A 1 PRACE 2IP RI 283493 74 23 08 2013 D10 2 Second Annual Report of WP10 A 100 files of IGB A 100GB A 4 A 100 files of IGB A 100GB B 8 A 100 files of 1GB A 100GB C 16 A 100 files of 5GB B 500GB A 4 A 100 files of SGB B 500GB B 8 A 100 files of SGB __ B 500GB C 16 9 A 100 files of 10GB
4. A host system with a GNU Linux operating system supporting TCP buffer auto tuning must be used Auto tuning technique allows receiver buffer size and TCP window size to be dynamically updated for each connection maximizing the action of a congestion algorithm which is recommended to be cubic or htcp as documented here https prace wik1 fz juelich de bin view Prace2IP Wp10 Task2 FileTransferTechnoSub taskActivities http fasterdata es net host tuning linux expert PRACE 2IP RI 283493 45 23 08 2013 D10 2 Second Annual Report of WP10 Recent versions of Linux version 2 6 17 and later support auto tuning with a default maximum value for the TCP buffer size of AMByte 4194304 bytes e memory reserved for TCP receiver buffers user sender host sysctl net ipv4 tcp rmem net ipv4 tcp rmem 4096 87380 4194304 e memory reserved for TCP sender buffers user sender host sysctl net ipv4 tcp wmem net ipv4 tcp wmem 4096 16384 4194304 It is suggested to increment the maximum value for both sender and receiver buffers depending from the network card and the BDP measured The following references help to check whether the maximum TCP buffer size is coherent with the measured BDP As example for a host equipped with 10G NIC and RTT delay below 100ms is preferable to set a value greater than 4MB 16MB or 32MB http fasterdata es net host tuning linux http www psc edu index php networking 64 tcp tune MTU and Jumbo
5. Second Annual Report of WP10 Mega 10 Hertz frequency 10 periods or clock cycles per second Mega 10 operations per second usually integer or logic operations Memorandum of Understanding Max Planck Gesellschaft Germany Message Passing Interface Maximum Transmission Unit Network File System Network Interface Controller Nemzeti Inform ci s Infrastrukt ra Fejleszt si Int zet National Information Infrastructure Development Institute Hungary National Supercomputing Centre in Link ping Sweden Open Graphic Library Operating System Pluggable Authentication Modules Pixel Compression over Internet Protocol Peripheral Component Interconnect express also PCI Express originally Personal Home Page now Hypertext Preprocessor Persistent IDentifier Parallel Network File System Portable OS Interface for Unix Partnership for Advanced Computing in Europe Project Acronym Poznan Supercomputing and Networking Centre Poland Redundant Array of Independent Disks REpresentational State Transfer Round Trip Time Rechenzentrum Garching der MPG Germany Storage Area Network Serial Attached SCSI Serial Advanced Technology Attachment bus Silicon Graphics Inc Swedish National Infrastructure for Computing Sweden Swedish National Infrastructure for Computing Link ping University Syst me de Publication pour l Internet Partag Solid State Disk or Drive Security Token Service Dutch national High Perform
6. ext4 the random read performance can be lowered by a factor of 3 or in extreme cases even 10 compared to a raw block device PRACE 2IP RI 283493 22 23 08 2013 D10 2 Second Annual Report of WP10 for synchronized and also asynchronized operations alike The reason for this is most likely a lot of inode related operations before accessing any data part of a selected file The random write performance is mostly not affected and its speed on an ext4 file system is about 60 80 of the speed of the raw block device for synchronized as well as for asynchronized operations Raw block device versus ext4 performance 18 000 15 734 16 000 14 000 12 000 10 000 8 000 o Ul up I N a Ui io co Ul 6 000 E 4 000 2 000 6 a E d m C o of E raw block device bandwith KB s B ext4 bandwidth KB s Figure 2 File System Performance Comparision Raw Blocks In most cases introducing a distributed file system to the setup does not further decrease the random read performance but in fact in most cases even raises it by a factor of about 2 since the data can be retrieved from two locations servers This is a perfect example of the read performance benefitting from a distributed file system This advantage turns to the opposite when looking at the random write performance of a distributed file system It is slowed down by a factor of 5 to 9 compared to the ext4 file system This drawback most likely can be reliev
7. it is planned to continue the work in the task T6 3 of PRACE 3 IP which is scouting technological developments of potential interest for PRACE Thus Gfarm and Coda should be made working properly for testing their behaviour XtreemFS and further ones like OrangeFS and FraunhoferFS missed in the search process should be included in an authoritative comparision Furthermore the scalability and multi site usage should be tested with more partners on the PRACE distributed infrastructure 4 Remote Visualization 4 1 Introduction As stated in deliverable D10 1 3 the work regarding remote visualization solutions systems and services has mainly focused on the class of solution that are application transparent as much as possible and session oriented so each users own their visualization sessions Those solutions are mainly represented by VNC like systems Among the different available VNC solutions reported in the previous deliverable PRACE centres have relied on Turbo VNC VirtualGL open source solution for deploying visualization PRACE 2IP RI 283493 27 23 08 2013 D10 2 Second Annual Report of WP10 services over WAN offering remote visualization services even at researchers at home connected with consumer grade ADSL lines Each partner has organized its visualization service using different hardware and adopting different access policies queued sessions advanced reservations special reserved visualization nodes but all
8. mount process freezes 6 6 5 iRODS Performance Evaluation Introduction The goal of this document is to evaluate the file transfer performance aspect using the 1RODS tool in a test environment Only the throughput aspect of the performance is studied Purpose As described in 2 IRODS is a data grid software system providing access to storage distributed on multiple sites and heterogeneous hardware and software storage Several aspects of this tool could have been studied but this report focus on the performance part PRACE already offers two services 5 to users to transfer data across the infrastructure The first one is GridFTP which 1s a data transfer tool defined as a core service and the second one is GPFS MC which is a distributed filesystem defined as optional The purpose of this study is to evaluate the IRODS protocol which comes with the 1RODS data management tool Performance analysis based on various tools has been achieved in the New File Transfer Tool task 1RODS wasn t evaluated in this scope as it is not a file transfer tool only and provides also a rich and large additional number of data management functionalities far over a simple file transfer tool Therefore it is studied as a separate tool in the 1RODS task The major user concern when using a file transfer service is the data access time when data is not located on the site where the user wants to use them So the performance to access user data
9. support and maintenance of components of the Globus Toolkit 31 including GridFTP and GSI SSH which are currently deployed in PRACE A MoU which describes the activities of the collaboration was signed earli this year For the evaluation of the GlobusOnLine tool IGE provided feedback on questions and problems Another important activity 1s the support for the Globus tools in production by PRACE The IGE project ended March 2013 however IGE support activities are continued by the European Globus Community Forum EGCF 32 2 4 EGI With the European Grid Infrastructure EGI 33 besides the interoperation of the helpdesks also the exchange of resource usage information was discussed This will provide user communities that use resources in different infrastructures a single view of their resource usage and can help users in choosing the most appropriate resource to run their jobs The technical details to enable such an exchange have been discussed between PRACE and EGI however the implementation 1s waiting on efforts by EGI 2 5 Pilots with user communities on data requirements A new activity named Data sharing and uniform data access across e infrastructures and community centres between PRACE EGI and EUDAT 34 started this period to address issues of data management interoperability The objective was to identify use cases of user communities that need to share data among these three infrastructures and to identify limitations an
10. tested and successful deployed to match the new conditions PRACE 2IP RI 283493 8 23 08 2013 D10 2 Second Annual Report of WP10 Furthermore to cover the complete production environment of PRACE several new Inca reporters for different middleware tools were developed These include the version tests for the prace service script and configuration which are an essential part of the middleware services Further Inca reporters testing for the existence and version number of the gtransfer tool the myproxy chent the GSI SSH client and the GridFTP client have been developed They are either based on existing reporters or have been developed from scratch AII mentioned Inca reporters have been tested and deployed successfully In addition a new Inca reporter prototype for the PRACE accounting infrastructure based on DART 28 was developed It is currently in testing stage and will be transferred into production soon It is currently evaluated if it may serve as a template for monitoring the Grid SAFE based accounting as well 2 7 Collaboration with other technological oriented projects since the beginning of the project PRACE has actively collaborating with other e Infrastructures and EU project to improve users experience strengthen the collaboration with external technology providers exchange knowledge among technical people raise the awareness around PRACE services and disseminate its activities Over the course of the second year of th
11. 2 Will data be shared among Projects members 9 3 Should data be shared with other research infrastructures and which ones 9 4 What is the size of data to be shared with others partners 9 5 For how long do you need to share your data with others partners 9 6 Is public access to the data needed PRACE 2IP RI 283493 43 23 08 2013 D10 2 Second Annual Report of WP10 10 Data sharing technologies 10 1 What technology service do you use for sharing 1f this sounds too technical to you just don t answer 10 2 Are you interested in advanced interfaces for sharing your data e g web portals 10 3 Which kind of tool or service would you prefer to use 1n the future 10 4 Are there different sharing needs during a project period compared to the time in between project periods PRACE DECI calls e g private data that may become publicly available Data Post Processing 11 Post processing and data analysis 11 1 Are you doing or planning to do post processing using PRACE resources 11 2 If yes which tools are you familiar with and using 11 3 Could you reduce the need of data movement if you could post process inside PRACE 11 4 Do you have needs for remote visualization during or after your jobs 11 5 Are you investigating on novel data analysis approaches using Map Reduce or NoSQL Other 12 Are there sufficient guidelines available on how you should deal with data in PRACE 13 Would you like to have best practices guides for r
12. 92 8 24 62 75 69 72 UniGine 1024x768 on rvn Cineca LAN Windows client client compression server cpu client cpu client cpu server client PLINK vncviewer fps fps lossless zlib 50 15 79 9 8 9 2 4 jpeg max quality LAN 60 18 15 7 12 6 9 jpeg med quality default 70 10 15 7 12 6 9 jpeg low quality WAN 60 10 12 7 12 7 9 UniGine 1024x768 on visual Cineca LAN Windows client client compression sshd tunnel server client cpu client cpu server client login cpu ei ser E eat ep Si tot fps fps lossless zlib 8 40 20 12 79 25 60 2 ipeg max quality LAN 20 40 80 35 26 25 35 20 32 jpeg med quality default 25 35 67 26 29 35 50 37 42 jpeg low quality WAN 6 1096 65 17 22 40 50 41 42 UniGine 1024x768 on visual Cineca LAN Linux client client compression sshd tunnel server client cpu client cpu server client PRACE 2IP RI 283493 51 23 08 2013 D10 2 Second Annual Report of WP10 login cpu cpu ssh vncviewer lossless zlib l 1 159 22 27 12 16 8 12 jpeg max quality LAN 18 2096 53 7699 32 44 36 56 25 40 21 31 jpeg med quality default 14 23 67 729 24 2896 36 48 jpeg low quality WAN 3 7 35 43 12 16 28 32 UniGine 1024x768 on visual low speed ADSL Windows client client compression sshd tunnel server client cpu client cpu server client login cpu Yo cpu Yo PLINK fps fps vncviewer j
13. Another approach is to use a graphic application that 1s able to render enough frames to saturate the image transport layer but 1s nevertheless representative of a real application with sufficient image complexity and variance For that purpose a demo of a graphics engine that pushes the limits of our old GPU but run smooth on new ones has been executed The tests have confirmed that the default settings that TurboVNC defines for the image compression setup are indeed the most appropriate for LAN as well as for high speed WAN as with them TurboVNC exhibits very few compression artifacts almost unnoticeable and optimizes all other costs as well as frame rate Depending on available bandwidth it could be necessary to adopt more aggressive image compression settings in order to make use of the full GPU power available to attain a perceptual satisfactory experience The next figure shows from left to right the same image as a sequence using lossless zlib lossless jpeg and default settings there is almost no noticeable artifact Figure 4 Images compressed with lossless zlib lossless jgeg and default settings While the next figure again from left to right shows the sequence with jpeg compression suggested for WAN custom compression set to 12 and custom compression set to 7 PRACE 2IP RI 283493 31 23 08 2013 D10 2 Second Annual Report of WP10 The two latter compression factors cause really annoying artifacts Thus testing was lim
14. C 1000GB A 4 10 A 100 files of 10GB C 1000GB B 8 _ A 100 files of 10GB C 1000GB C 16 B 1 file of 100GB A 100GB fA 4 B 1 file of 100GB _ A 100GB B S 100GB C 16 500GB 4 500GB B 8 500GB C 16 3 o TOE in UJ UJ UJ UJ UJ i UJ UJ Ne B 1 file of 1TB C 1000GB A 4 B 1 file of ITB C 1000GB B 8 B 1 file of ITB C 1000GB C 16 CATT N I LT WOjooj oc0 muniu h2 General Network Information CINECA Gbps 4194304 bytes 4194304 bytes IDRIS 1 Gbps 6291456 bytes 4194304 bytes 1 Gbps 4194304 bytes 4194304 bytes This dataset case is able to provide a complete and detailed set of cases to analyze the performance of transfer tools but with our test bed we were not able to use it because of the following three reasons e The main one is only CINES provided an access to the dedicated high performance 10Gbe PRACE network All other sites can only provide a public internet access to their IRODS server limited to a 1 Gbe shared link e Only CINES provided a gridFTP environment on the 1RODS server to perform the reference measures e IDRIS and NIIF couldn t provide enough disk I O performances on the test IRODS server they offered for the test Actually the tests that were performed are the following ones PRACE 2IP RI 283493 75 23 08 2013 D10 2 Second Annual Report of WP10
15. PRACE 2IP RI 283493 T 23 08 2013 D10 2 Second Annual Report of WP10 Purpose The 1RODS organizes its storage locations as resources The Direct Access resources feature provides support for sharing access to a storage location i e resource with the regular filesystem access A typical usage scenario would be an environment in which there is a shared high performance file system mounted on a compute cluster via NFS and on which iRODS has the files from this file system registered in order to provide meta data annotation for the files in this file system 1 e IRODS acts as an overlay for the UNIX file system To make this possible the system relies on identical user ids and passwords on the 1RODS service and user clients Also this feature relies on the filesystem metadata feature being enabled during 1RODS installation This evaluation will test the basic functionality of creating the direct access resources and accessing the files comparing what is seen through 1RODS access with direct filesystem access Responsibility SITE ROLE TASK Full evaluation References List the references applicable documents and related documentation user technical 1 https www irods org index php Release Notes 3 2 2 https www irods org index php Direct Access Resources 3 https www irods org index php File System Meta data 4 https www irods org index php Run server as root 5 https groups google com forum topic
16. RI 283493 l 23 08 2013 D10 2 Second Annual Report of WP10 1 Introduction The objectives of WP10 are e Enhancing the existing Tier 1 operational infrastructure e Evaluation of additional data services e Remote Visualization Each of these objectives has a corresponding task in the work package Where appropriate the tasks are organised in sub tasks to better focus the specific topic Structure of the Document The following document consists of three further chapters Enhancing the Existing Infrastucture Evaluating Data Services and Remote Visualization one for each of the tasks addressing one of the objectives listed above The single chapters then contain several sections covering the work of the respective sub tasks which are logically mainly independent from each other A chapter Summary and Future Work will conclude and finally an Appendix with several sections provides even more detailed or additional information for some of the tasks or sub tasks Relation to WP6 Operations in PRACE WPO is responsible for the operation of the infrastructure of and the services provided in PRACE As in PRACE IIP again in PRACE 3IP the technological evolution is covered as task T6 3 of WP6 while in PRACE 2IP the separate work package WP10 was dealing with technological developments The deep collaboration between WP10 and T6 3 1s achieved by having joint bi weekly video conferences coordinating the work 2 Enhancing the Existing Infras
17. Training events EEUU ears nema Sree a gap men ee PE eee DES ANE EL CED ES E vr ITUNES VICES ONSE EE OS Kei RE eT TE FAD PRACE Training events in a Global Race Duganizaton intres B ai Hints The PRACE RI is open to all European researchers affiliated with recognized announcement European academic institutions and industries pips The aim of PRACE RI Training is to provide a sustained high quality training G Call Announcement and education service for the European HPC community through seasonal A I vacancies Industry Access schools workshops and scientific and industrial seminars in order to PRACE Pratatypes effectively exploit the unprecedented capabilities of HPC resources of RI c PRACE PRACE Resources Guiding principle in developing the training and educational programs is the p reete How io apply compliance with the educational needs of the researchers thorough za PRACE Awarded understanding of their existing skills and competencies atl PRACE Prajecus Newsletters PRACE Peer Rewiew a PRACE Training Fortal Home WATCH LIVE HOW SCHOOL 2013 rek PRACE hnpiermeninton Phares Code coupling using OpeniPaim d AE A Statistical Analysis Tor Misi Post Gereoam c Date AUTUMN PRACE Training Portal Parallel Computing with R SCHOL 2013 Supmmeer ot HPC zn 23 27 September Hace digi ub ee PRACE Advanced 4 4 45 E Uubliana Slovenia Training Cenires n ba
18. checklists includes e Uniform access to HPC partial PRACE internal interactive command line access to HPC complete Data transfer storage and sharing complete Authentication partial Authorization partial Accounting partial Grid SAFE Accounting repository complete Network management partial Monitoring partial Software Management and Common Production Environment complete The main implementation details behind Service Certification have been decided First of all certification results should be stored on the wiki using special templates called the Service Certification Log The quality checklists will be continuously improved by the sites assigned to perform the specific certifications Finally it was decided that certification results will be internal for PRACE staff only The activity will be continued in Task 6 3 of PRACE 3IP and will include performing and improving the certification procedures and integration of selected service certification tests with INCA for procedure automation PRACE 2IP RI 283493 5 23 08 2013 D10 2 Second Annual Report of WP10 2 4 DECI Portal The objective of this activity was to setup a tool to improve the submission and management of the project proposals in the DECI calls 20 A steering committee was set up and comparing the PRACE Tier 0 PPR developed by CINES against the HPC Europa PPR developed by CINECA See appendix 6 2 1 for the functionalities comparison table It w
19. defining a common methodology next steps have been the selection of a set of tools and the set up of test benches The activity has performed already several tests but it 1s too early to provide a summary and conclusion about the results this will be done after all tests have been finalized in task 6 3 of PRACE 3IP Then a separate document will be produced The complete current description of the methodology including all its definitions can be found in the appendix 6 4 while the next section cites relevant parts of the appendix 6 4 4 describing the methodology 3 2 1 Common Methodology A common methodology for evaluating new file transfer technologies where new stands for not officially supported in PRACE has been designed to be independent from a specific software solution Similar work carried out in other scientific contexts has been also taken into account to ensure robustness and completeness in the way of making tests 14 The following factors have been considered as essential features of the methodology e Assessments must be produced in a consistent manner across different sites and different network paths e Production conditions and any network turbulence must be considered e Performance must be measured with different types of workloads and different numbers of parallel streams only tools supporting parallel data streams must be considered e Atemplate must be available for collecting results of tests e A me
20. easily The sub task New File Transfer Technologies compared several file transfer tools currently not yet supported by PRACE in the service catalog and assessed their potential benefits for users in PRACE Initially the iRODS repository technology had been considered the most promising software concerning handling of data identified by persistent identifiers and was therefore included into the software to be evaluated by WP10 As the pilot collaboration with EUDAT showed there is a real user need for such a technology Therefore it has been evaluated very deeply Depending on further user or community requests IRODS may become a generally supported service at least optional in PRACE Finally several file system technologies have been investigated with respect to their possible use in PRACE The performance measured and the stability tested does not yet allow recommendations of any of the evaluted file system techonologies The third task Remote Visualization continued its investigations in different technological implementations based on VNC technologies The Remote Connection Manager pilot installation has been finalized and will become a production service in PRACE Furthermore performance has been tested for varying hardware and network configurations with different software implemenations being able to give recommendations on the best setup and usage of VNC based remote virtualization depending on the infrastructural conditions PRACE 2IP
21. enu words FAD Organisat Sinies t ndzBcon Hemberg CaS Anmaurncemeenti Industry Access PRACE Prototypes PRACE Feznamcez How do appiy PRACE Awarded Projects PARADE Peer Hiewiew SHAPE Plaid Public Deireeraties a PRACE impiemeninbon _ Phases DECI Ter l Access PRACE Training Parini Summer oT HPC PRACE AId vanced Training Centres P amp TCs PRACE Training event Bez Practor Guides PRACE User Deociumenimian Contkerences and mesinga PRACE industial Sears PRACE Outeach Events HPC Related Events i Prec Meiers Viena Nez Pubcon Preseninbons aievwens DON THE WEB Upcoming PATC courses Hn Meverits pesce d exicategory Dolan py catecid 2 Europe t plc naGlobal Race Advanced Fortran Topics amp Coarray Fortran CSC Monday 10 June 2013 Thes brand new course gives introduction to more modem features of the Fortran 2008 Standard Among subjects are more flexible handing of dynamically allocated objects 7 operator overoading object onented features and language interoperability Thereafter a los highly useful introduction to fe Cray Advanced Tools workshop EPCC PRACE Wednesday 26 June 2013 Press releases Cray s supercomputer platforms are an advanced paimng of software and hardware that provide HPC appbcation developers and users the opportunity of excellent scaling and PRAE high productivity This workshop provided staff from the Cray Cen
22. for DECI is considered to be at an early stage and the tool 1s still under evaluation together with WP2 Because of its relevance it will be continued in the technology task T6 3 of PRACE 3IP As of July 2013 the tool provides an implementation of a DECI peer review process with complete on line handling of proposals from the submission to the technical and scientific review assessment e Electronic submission of HPC project proposals o Incremental edition of proposals while a call is open o Declaration of project investigators collaborators o Final submission with prior checks for required fields being filled in e Follow up of the peer review process of a given call is mainly achieved from a master spreadsheet page where relevant data 1s display on the status of proposals and reviews e Technical evaluation TE where all relevant data from proposals are visible in TE form e Scientific Evaluation SE where evaluators can get limited access to relevant proposals and TE e Having on line forms allows for leveraging typical database abilities in terms of data extraction such as obtaining a list of proposals that match criterion or a list of registered users e Transverse functionalities notably encompass data export to Excel spreadsheets and PDF Interfaces to other PRACE services will be going to be implemented step by step into the PPR tool as part of T6 3 of PRACE 3IP e Interfacing with the DPMDB tool for project follow up e Interfacing
23. level i e evaluators can gain limited access to relevant proposals and TE This would cause different log in views for Applicants Technical amp Scientific evaluators and DAAC staff 6 Store applicants data project Essential Not Well HPC Europa Integration data TE review data suggested supported supported with the DECI database is extra TE info SE data ranking needed info etc into the DECI Database 7 Create and or change user s Essential Partialy Partialy PRACE TO Planned evaluator s site s countries supported supported 2H2012 and or portable info kernel roadmap HPC Europa evaluators cannot change their info autonomously while users can 8 Support the process of Essential Well Well submitting a short report from supported X supported the PI after the completion of the project the template of this report being downloadable from the tool 9 Create statistics reports Desiderata Partially Well PRACE TO Should be ok of the DECI process 1 e number supported supported to be precised of technical evaluations per site number of scientific evaluations HPC Europa general per evaluator Moreover the statistics on the entire publications related to work done review process are with DECI resources should be available though tracked via the proposed tool 10 Copy or link the relevant data Essential Not partially PRACE TO Linking should from the web based tool when supported supported be possible
24. month day or in a calendar view Exporting Data InDiCo provides several ways to export data e To Personal Scheduler Tools Outlook 1Cal e RSS feeds e Sharepoint e HTTP Export API HTTP Export API InDiCo allows for programmatically access to the content of its database by exposing information like category contents events rooms and room bookings through a web service through the HTTP Export API The basic URL looks like http my indico server export WHAT LOC G ID TYPE PARAMS amp ak K EY amp timestamp TS amp signature SIG where e WHAT is the element to export one of categ event room reservation e LOC is the location of the element s specified by ID and only used for certain elements ID 1s the ID of the element to export can be a separated list TYPE is the output format one of json jsonp xml html ics atom bin PARAMS are various parameters affecting filtering sorting the result list KEY TS SIG are part of the API Authentication Details about HTTP Export API URL parameters can be found in the user manual PRACE 2IP RI 283493 4 23 08 2013 D10 2 Second Annual Report of WP10 2 2 8 SPIP Plugins and Features Syndication CMS built in feature The syndication system allows sharing the attached document urls podcasting transferring keywords tags from one site to the other as well as transferring the section or category of the articles The default templates provided by SPIP inc
25. news A cnaordinated approach tu science in Europe 19 June 20035 PRACE Educmiez TERITIEHTTPAV Ss Riexzenancdierz 23 08 2013 D10 2 6 2 DECI Portal Second Annual Report of WP10 The next picture shows a screenshot of the PPR Tool with a setup for the latest DECT 11 call after logging as applicant Partnership for Advanced Computing in Europe Frontpage DECI 11th New proposal Online application form Your proposal will be created upon clicking the CREATE PROPOSAL button at the bottom of this page It is advised that you fill in some minimal information by now and create your proposal before proceeding with the main content Please note that mandatory fields for final submission are indicated by a red square m Please complete all the required fields in this form and attach a 3 4 page summary of your project using the template provided This summary can be attached from link in the frontpage General information Type of proposal Start date Project title DECH11 Prosect November 1st 2013 Project acronym a FHPNS Astro Sciences Research field Contact person for all correspondance If different from the project leader below The project leader and the contact person will both receive all information Future proof High Performance Numerical Simulation for CFD with FEASTFLOW 2 Figure 9 DECI PPR Tool Screenshot 6 2 1 DECI peer review tools functionnality comparison ta
26. niifitest A admin Test procedure N Is a irods A iquest sql ShowPPtimes niifitest tempZone N ils PRACE 2IP RI 283493 61 23 08 2013 D10 2 Second Annual Report of WP10 Expected result There is no irodsA file present on client side There 1s no PAM derived password present in iCAT The ils command should not work It is expected that IRODS will prompt for IRODS password and then fail because this user does not have an 1RODS password Is a irods rodsEnv iquest sql ShowPPtimes niifitest tempZone No rows found ils Enter your current IRODS password rcAuthResponse failed with error 826000 CAT INVALID AUTHENTICATION Test result The test result matches the expected result Detailed tests PAM derived password does not exist 2 Inr conditions see PAM derived password does not exist 1 username that is niifitest zone name that is tempZone PAM asks for the password N n fitest A admin Parameters and input data Test procedure N iinit type invalid password Is a irods A iquest sql ShowPPtimes niifitest tempZone Expected result The iinit command should prompt for PAM password It should fail upon entering an invalid password As the user failed to authenticate a PAM derived password should not be present neither on client side nor in 1CAT iinit Enter your current PAM system password rcPamAuthRequest failed with error 993000 PAM AUTH PASSWORD F
27. not include only data rate but also reliability This means that work started in DEISA and followed 1n PRACE 7 8 for improving user interaction with GridFTP and spreading its use are indeed really appropriate Tests of GlobusOnLine confirmed also excellent responses on this direction PRACE 2IP RI 283493 17 23 08 2013 D10 2 Second Annual Report of WP10 Dedicated hosts for Data Transfers In order to provide a reliable sustainable and high performance data transfer service it is suggested to consider dedicating specific systems for inbound and outbound data movements This 1s because specific host configurations can affect other production services and also because the amount of data to be transferred 1s even more huge and it really needs dedicated hardware for load balancing This feedback mainly addresses a common PRACE strategy on big data c f chapter 3 1 More tests are needed It is strongly suggested to make a follow up activity in PRACE 3IP by reusing the same methodology presented here and all efforts spent in setting up the different test benches It is recommended moreover to extend tests by allowing all involved partners to test all tools against GridFTP 3 3 iRODS integrated Rule Oriented Data System This sub task follows the 1RODS evaluation initiated in the DEISA project Its scope was to evaluate the recent tool enhancements and to assess the current user needs To achieve its goals the work was split into two sepa
28. not work with an invalidated PAM derived password It is expected that iRODS will prompt for an iRODS password and fail because the user has no iRODS password Test result 205 Enter your current IRODS password rcAuthResponse failed with error 826000 CAT INVALID AUTHENTICATION PRACE 2IP RI 283493 66 23 08 2013 D10 2 Second Annual Report of WP10 EN The test result matches the expected result Detailed tests An existing PAM derived password is invalidated on client side 2 Initial conditions see PAM derived password does exist 1 rodsA is invalidated timestamp changed Parameters and input data rodsA at client side Test procedure touch m t lt original timestamp gt irods irodsA ils Expected result The ils command should not work again if the invalidated PAM derived password s timestamp is changed back to the original value Test result After touch command ils Enter your current IRODS password rcAuthResponse failed with error 826000 CAT INVALID AUTHENTICATION The test result matches the expected result Detailed tests An existing PAM derived password is invalidated on client side 3 eL E e PAM E 2 initial conditions see derived password does exis PAM asks for the password Parameters and input data Test procedure Expected result The ils command should work again after re authentication iinit Test result ils Enter your current IRODS password rcAuthRespons
29. of limited search and authentication options On the other hand the 1Drop Web Interface is fast intuitive and easy to use It also shares limited search options and authentication problem but it broadens its options by including a direct link to the 1Drop Desktop GUI to complement some of its flaws To get the most of the 1Drop functionality Desktop and Web interface should be used together Hopefully further development and future versions will make this GUI for 1RODS an obvious choice for both end users and administrators but for the time being mostly due to lack of support for GSI its use 1s limited PRACE 2IP RI 283493 87 23 08 2013
30. used the same underlying technological platform using the VirtualGL project for application neutral OpenGL remotization scheme and Turbo VNC as the VNC server client component SURFsara has investigated a number of remote visualization topics that are of interest due to current trends in computing and visualization A first topic was the use of VirtualGL TurboVNC for high end high resolution large screen visualization setups Secondly an investigation was made into the possibilites of using GPU compute hardware for remote visualisation together with a comparison between GPU based rendering and software rendering The detailed results will be published in a separate PRACE whitepaper CINECA had used a proprietary VNC technology from IBM DCV to support technical users that need specific proprietary visualization applications in engineering and flow simulation StarCCM Fluent etc The DCV technology is currently provided and supported by NICE and is still in use as an embedded component of a customized web portal for access to technical computing resources based on NICE EngineFrame SNIC LiU explored other remote visualization technologies available and investigated deeply into the Teradici PcoIP solution described below in section 4 2 It can be used when top performance or complete application transparency were needed and a high speed low latency campus wide network backbone was available The second year focused on the evaluation of the per
31. with dependencies jar transfer applet for embedding in iDrop Web Interface Tested Features iDrop desktop and web interfaces Non Tested Features Other features of IRODS are not part of this evaluation PRACE 2IP RI 283493 83 23 08 2013 D10 2 Second Annual Report of WP10 Tests Phases e Phase 1 included setting up the testing environment and enablement of iDrop Web Interface service e Phase 2 included exploration and testing of iDrop features and functionality Tests Environment SITE IPB IRODS server KVM Virtual Machine on PARADOX Scientific Linux 6 3 2 Proc Intel Xeon CPU E5345 2 33GHz 2 GB RAM 1Gb Ethernet card Client Microsoft Windows 7 Home Premium Ubuntu 12 04 Java SE Update 25 on both OSes Apache Tomcat 6 0 on both OSes iDrop explicitly requires username and password for authentication so it is important to note that iDrop doesn t work with GSI authentication Therefore this testing was performed on an iRods 3 2 instance without GSI support Testing Methodology Evaluation was performed as a series of feature tests following the test script adopted and revised from the 1Drop project home site https code renci oreg ef project irodsidrop wiki pagename iDrop testing script The goal of this evaluation is to get familiar with iDrops features and to examine its functionality ease of use reliability and efficiency Tests Description For each group of functional features descriptio
32. with quite PRACE 2lP RI 283493 37 23 08 2013 D10 2 Second Annual Report of WP10 Functionality Rate HPC Europa Tool PRACE Tier 0 PPR Tool Comment needed into the DPMDB ze project name home site technical requirements such as CPU type number of jobs memory simulation codes etc reasonnable work HPC Europa Integration with the DPMDB is needed 11 Copy summary of projects resource usage from DPMDB to the web based tool so that PIs can view accounting information related to their projects without learning a new tool DART Desiderata Not supported Partially supported PRACE T0 Should be possible 12 Create and export documents and information that should feed other systems or processes i e automatic generation and export of PDF s for mailing at any point in time Enable generic export all documents related to a call to be exportable in corresponding folders files e g one folder Astrophysics containing as many as folders as proposals each containing all the documents related to this proposal application tech review scientific review Desiderata Partially supported Well supported PRACE TO Such features already exist in PPR tool but some specific development may be necessary to fit the requirements HPC Europa Most of the information can be easily exported via Excel file format nor PDF 13 Keep extensive logs regar
33. 0 Collaboration with providers of other data storage like national storage systems community storage and project storage and providing fast connectivity to them could improve transfer speeds for data movements in and out of PRACE internal storage systems Finally PRACE support could provide users with more specific help for individual needs concerning data transfers to and from PRACE systems The realizations of such options require respective decisions on the executive level of PRACE followed by the implementation on the operational level Other more long term Options More effort 1s required for some of the other possible options which sometimes require even changes in the policies and service provisioning models e Access to analysis and visualization nodes closely coupled with the HPC systems The need for data movement can be reduced 1f not avoided and the workload for the users decreases dramatically for some projects c f Remote Visualization in 4 e Availability of intermediate preservation of data for volumes and sharing With the provisioning of intermediate storage data can be shared more easily between sites and big volumes can be handled cheaper The intermediate storage can be handled in a centralized way with decentralized resources cf File System Technologies in 3 2 e Provision of technologies like IRODS cf 3 3 also for intermediate storage Conclusions It is evident that PRACE requires a clear strate
34. 10 2 Second Annual Report of WP10 Adding metadata to file in direct access resource through iRODS and local filesystem e Direct access resource at path opt rodshare e adummy text file testfile txt used to verify that resource is accessible for file operations e Metadata triplet added has following details o AttrName TextType o AttrValue random o AttrUnit none Test procedure e iinit type in prl1g000 s password e imeta add d testfile txt TextType random Expected result The test file should have metadata record associated with it Parameters and input data imeta Is d testfile txt AVUs defined for dataObj testfile txt attribute TextType value random units The test result matches the expected result Test result Direct access resource creation on NFS e See 2 Direct access resource file creation through iRODS and local filesystem nfs mounted on nfs e Parameters and input data e Direct access resource at path nfs e The resource created is named nfstestresc iinit enter rodsadmin s password iadmin mkresc ntfsresc direct access filesystem cache irods ipb ac rs nfs nfstestresc resource should be created on the irods Expected result Instance ilsresc demoResc testresc nfstestresc The test result matches the expected result Initial conditions lest procedure i e Test result File access and creation through iRODS and filesystem in direct access resour
35. 1irods org 1index php glossary 4 https www irods org index php Downloads 5 https bscw zam kfa juelich de bscw bscw cgi 942629 6 http fasterdata es net host tuning linux 7 http www iozone org Tested Components The transfer bandwidth performance evaluation focus on the IRODS software itself but also on the network 6 and I O environment involved 7 The related documentation on iRODS is specified at 1 2 3 The iRODS release used for the evaluation is 3 2 It can be downloaded at 4 Tested Features The transfer bandwidth performance evaluation will focus on 3 major points a The test case choice of the file and process b The testbed servers involved iRODS servers network configuration c The tests The methodology is the most important part of this work This way relevant results can be produced and compared Tests Environment SITE TEST ENVIRONMENT SGI Altix XE 250 e 2x Intel E5420 2 5 GHz e 322GB RAM PRACE 2IP RI 283493 73 23 08 2013 D10 2 Second Annual Report of WP10 e Gb Ethernet card e 10 Gb Ethernet card e SUSE SLES 11 SP1 Local FS Lustre 577 MB s R W iRODS 3 2 DELL CINECA e 16x Intel E5530 2 4 GHz e 64GB RAM e Gb Ethernet card e Debian GNU linux 6 0 Local FS GPFS NFS 160 MB s R W iRODS 3 2 IBM System x3655 IDRIS e 4x AMD Opteron 2218 2 6 GHz e 2GBRAM e 1Gb Ethernet card e Fedora release 17 Beefy Miracle Local FS ext4
36. 1s an important question PRACE is a high end HPC infrastructure in Europe The data used on this infrastructure is in the same order as the compute power it provides We are talking here about Terabytes of data manipulated across the infrastructure At this scale the number of files cannot be the most important element comparing to the volume although it can be also an issue iRODS can be setup to allow a cross access from several sites providing themselves their own iRODS server In this case servers are interconnected thru remote zones where accesses are restricted to authorized users PRACE 2IP RI 283493 T2 23 08 2013 D10 2 Second Annual Report of WP10 The evaluation consists in transferring files using different set of parameters for evaluating the transfer bandwidth obtained from the end user point of view Note that this evaluation relies on a testbed far from a production environment It is based on heterogeneous hardware at each site so has to be considered as a first step evaluation waiting for better network connections disk and systems when available at each site Responsability SITE ROLE TASK CINES Full evaluation CINECA IDRIS NIIF Support to setup and configure IRODS servers for the workbench References 1 https www irods org index php Release Notes 3 2 2 https www irods org index php IRODS Data Grids Digital Libraries Persistent Arch ives and Real time Data Systems 3 https www
37. 4 5 Test cases Fixed a medium which could be Internet or the private PRACE network and taking into account of the methodology above mentioned there will be 18 runs to execute for each specific tool The following table shows an example for two specific dataset types 100 files for Dataset A against 1 file for Dataset B amun Dataset Workosd Parallel Streams _ e AtWewsob Bewe cos e AtWeowG cuwee 80 o anonsas cuoce cos Table 9 File Transfer test dases with at least 18 runs each 6 4 6 Template for testing the data transfer tool Results must be collected by data sheets based on a predefined layout A data sheet will include quantitative data as well as information about the test bed used It acts as a data base from which structured information can be further elaborated e g performance with a fixed dataset type and different workloads and parallel streams performance with a fixed workload and different dataset type and parallel streams etc Information can be presented in table and or graphic format recommended General Information Tool SiteA S BBCP CINES EA Type Capacity BDP Internet 200Mbps 1250 KBytes Hosts configuration PRACE 2IP RI 283493 49 23 08 2013 D10 2 Second Annual Report of WP10 Max TCP Buffer Size Site A Max TCP Buffer Size Site B net ipv4 tcp rmem net ipv4 tcp wmem net ipv4 tcp rmem net ipv4 tco wmem 4194304 4194304 4194304 4194304 Quantitative Assessmen
38. AILED Is a irods JrodsEnv iquest sql ShowPPtimes niifitest tempZone No rows found Test result The test result matches the expected result Detailed tests PAM derived password does not exist 3 PRACE 2IP RI 283493 62 23 08 2013 D10 2 Second Annual Report of WP10 Initial conditions see PAM derived password does not exist 1 username that is nifitest zone name that 1s tempZone PAM asks for the password N niifitest A admin Parameters and input data Test procedure N iinit type valid password Is a irods A iquest sql ShowPPtimes niifitest tempZone Expected result The iinit command should prompt for PAM password It should succeed upon entering the valid password After successful authentication a PAM derived password should be present both on client side and in 1CAT Test result 5 init Enter your current PAM system password Is a irods ArodsA irodsEnv iquest sql ShowPPtimes niifitest tempZone 1209600 01372080475 01372080475 ShowPPtimes shows that expiry timestamp 1s two weeks 1209600 seconds As the PAM derived password is just created creation and modification timestamps are the same The test result matches the expected result Detailed tests A valid PAM derived password does exist 1 e see Basic PAM test PAM derived password exists both on client side and iniCAT Parameters and input data _ irodsA on client side Initial co
39. AM Authentication 3 https www irods org index php PAM LDAP Authentication Authorization 4 https www irods org index php PAM SSL Setup 5 https www irods org index php glossary 6 https www irods org index php Downloads PRACE 2IP RI 283493 59 23 08 2013 D10 2 Second Annual Report of WP10 Tested Components The PAM LDAP Authentication evaluation focus on the software The related documentation is specified at 1 2 3 4 The iRODS release used for the evaluation is 3 2 It can be downloaded at 6 Tested Features Evaluation will focus on PAM LDAP integration Non Tested Features Other features of IRODS are not part of this evaluation Tests Phases There are two phases of evaluation a Check if PAM works SIRODS HOME server bin PamAuthCheck will be used for this test b Test cases These tests include IRODS icommands and other IRODS related activities Tests Environment NIIFI 1RODS test server Cloud VM Debian 6 0 7 2 cores 0 5 vcpu i e 50 guaranteed cpu time 2GB RAM 1Gbit Ethernet NIIFI LDAP server Cloud VM Debian 6 0 7 l core 1 vcpu i e 100 guaranteed cpu time IGB RAM I Gbit Ethernet software OpenLDAP PAM configuration etc pam d irods for irods service auth required pam ldap so Testing Methodology The test will focus mainly on the functional aspect It will be checked whether the functionality 1s provided and it does what 1s supposed to d
40. ANT2 is the follow up as of 2004 Grand Equipement National de Calcul Intensif France Giga 10 Floating point operations usually in 64 bit i e DP per second also GF s Giga 10 Hertz frequency 10 periods or clock cycles per second Gigabit Ethernet also GbE GNU s not Unix a free OS Graphic Processing Unit Grid Security Infrastructure GSI based ssh secure shell remote login facility High Performance Computing Computing at a high performance level at any given time often used synonym with Supercomputing Formerly known as International Business Machines Institut du D veloppement et des Ressources en Informatique Scientifique represented in PRACE by GENCI France Initiative for Globus in Europe Integrated Digital Conference Input Output Institute of Physics Belgrade Integrated Rule Oriented Data System J lich Supercomputing Centre FZJ Germany Kilo 2 10 Bytes 8 bits also KByte Laurence Livermore National Laboratory Livermore California USA Leibniz Supercomputing Centre Garching Germany Multiscale Applications on European e Infrastructures Mega 2 10 Bytes 8 bits also MByte Mega 10 Bytes 8 bits per second also MByte s Mega 10 Floating point operations usually in 64 bit i e DP per second also MF s RI 283493 23 08 2013 Vill SNIC LIU SPIP SSD STS SURFsara TB TFlop s Tier O UNICORE USB VIPAVPH VM VNC XML PRACE 2lP
41. CE Taming Werkesdey 17 Jury 2013 Eun OpenMP is the industry standard for shared memory programming which enables senal programs to be parallelsed using compiler directry es This course amp aimed ai programmers seeking to deepen ther understanding of OpenMP and explore some of is more recent and advanced features This two day Advanced Fortran Topics LRZ Monday 15 September 2013 Thus course m targeted at scientists who wish to extend ther knowledge of Fortran beyond what rs provided in the Fortran 95 standard Some other tools relevant for Fortran 95 2003 CSC Monday 30 Sepeermber 2013 Figure 7 Event Integration Screenshot 2 Upcoming PATC Courses 34 23 08 2013 PRACE 2IP RI 283493 PARTNERSHIP FOR ADVANCED COMPUTING Ine 8 ste 3 Second Annual Report of WP10 Home page gt Training and Documentation PRACE Training events PATC Courses PRACE in a eu wonda FAQ Ongantsaton PATC Courses Sin hies Sintzbcs o oO June 2013 Tal Announcement Cote coupling using CipeenePn m d CINES industry Acoe zs PRACE Proiniypes Statistica Arshysis for Post Genaximic Daiki PARACE Hie zaamoes Haw ic apy i Parle Computing wiih A PRACE Awarded EPCC PRACE Peer Ae wee i SHAPE Pio Pubic Deityerntiies i PRACE mpiemeninion Phases a Hybrid MPUDpenMP programming IDRIS 25 27 Cray Advanced Tools workshop 4 EPCC T T p DECI Tier 1 Access i Training and Di
42. Ethernet Frames Ethernet s maximum frame size of 1500 bytes is not optimized for Gigabit Ethernet network cards and can actually inhibit the ability of applications to take full advantage of a high network capacity This limitation can be overcome by changing the MTU to a value of 9000 allowing Ethernet frames with a payload of 9000 bytes Assuming etho as the name of the network interface the MTU can be changed with the following command user sender host ifconfig ethO mtu 9000 Permanent changes take effect by modifying network configuration files dependently from the specific Linux distribution installed Disk performance Before to run any test it is absolutely required to check performance of the disks subsystem involved I O benchmarks like hdparm bonnie and iozone could be used to test performance of I O operations on the disk Network capacity Tests will be executed over both public Internet and private PRACE network For public Internet the only requirement is that the user end point is plugged to a network with the following minimum requirements e RTT below 70ms e 0 of packet lost e Jitter not above Ims For hosts connected to the internal PRACE network no minimum requirements are set To check if the auto tuning is active the file proc sys net ipvA tep moderate rcvbuf must be present and with value equal to 1 http www cyberciti biz fag centos rhel redhat fedora debian l
43. GB e Workload B 500GB e Workload C 1000GB 1TB Parallel Streams Only tools that support data transfer parallelism can be considered Choosing the number of parallel streams 1s not a simple task because performance could decrease with high number of streams It mainly depends from the memory availability at the end points Several studies have shown that in practice using between 4 and 8 streams are usually sufficient 16 streams may be useful only in case of bad performance found with 4 and 8 Above 16 1s basically wasting resources So it is recommended to run test with 3 different numbers of streams e Parallel Streams Configuration A 4 e Parallel Streams Configuration B 8 e Parallel Streams Configuration C 16 Qualitative Factors It has been considered as valuable to take into account also qualitative factors that are not strictly related to performance of a specific file transfer tool Factors like reliability are important for providing a complete feedback whether deciding to include a specific file transfer tool into data services for PRACE Evaluation could be provided by using a ranking from 1 really bad to 5 really good along with a short comment specifying the motivation of the mark Recommended factors to be considered are Reliability Footprint Intrusiveness Maintenance Fault Tolerance Code Maturity Community Acceptance PRACE 2IP RI 283493 48 23 08 2013 D10 2 Second Annual Report of WP10 6
44. IROD Chat Q3 MvceznE3E 6 https www irods org index php glossary 7 https www irods org index php Downloads Tested Components This document evaluates the Direct Access Resources feature of IRODS 3 2 Tested Features This evaluation focuses on the file access through 1RODS and directly through the file system which is provided by the Direct Access Resource feature 2 Non Tested Features All other 1RODS features that are not directly relevant for the evaluation of the Direct Access Resources Tests Phases There were two types of tests performed e checking if the direct access works for resources on the local filesystem e checking if the access also works for shared filesystem mounted through NFS PRACE 2IP RI 283493 78 23 08 2013 D10 2 Second Annual Report of WP10 For each of the test cases the following subtests were executed e testing if the resource creation is working e testing if the file creation and the access through the filesystem and the resource are working Tests Environment The 1RODS server and clients icommands had to be build with the following build flags enabled in config config mk e FILESYSTEM META 1 e RUN SERVER AS ROOT 1 e DIRECT ACCESS VAULT 1 as described in 2 3 and 4 The server was started as the root user while database default PostgreSQL that comes bundled with the installation had to be run as a regular user IRODS users had to have acco
45. NRS Centre National de la Recherche Scientifique CPU Central Processing Unit DANTE Delivery of Advanced Network Technology to Europe DART Distributed Accounting Record Tool PRACE 2IP RI 283493 vii 23 08 2013 D10 2 DCV DECI DEISA DoE DPMDB EGCF EMI EPCC EPSRC FIO FUSE FZJ GB Gb s GB s GCS GEANT GENCI GFlop s GHz GigE GNU GPU GSI GSI SSH HPC IBM IDRIS IGE InDiCo I O IPB IRODS JSC KB LLNL LRZ MAPPER MB MB s MFlop s PRACE 2lP Second Annual Report of WP10 Deep Computing Visualization IBM or Desktop Cloud Virtualization NICE Distributed Extreme Computing Initiative Distributed European Infrastructure for Supercomputing Applications EU project by leading national HPC centres United States Department of Energy DECI Project Management Database European Globus Community Forum European Middleware Initiative Edinburg Parallel Computing Centre represented in PRACE by EPSRC United Kingdom The Engineering and Physical Sciences Research Council United Kingdom Flexible I O Tester File System in User Space Forschungszentrum J lich Germany Giga 2 10 Bytes 8 bits also GByte Giga 10 bits per second also Gbit s Giga 10 Bytes 8 bits per second also GByte s Gauss Centre for Supercomputing Germany Collaboration between National Research and Education Networks to build a multi gigabit pan European network managed by DANTE GE
46. PCC UK BBCP CEA France CINES France EPCC UK PRACE Internet SNICINSC Sweden internet 1 10Gbps Table 2 Test benches for evaluating new file transfer technologies As mentioned a preliminary test phase has been carried out within the available timeframe A full test phase with comparisons could require several months and the development of script for automating tasks 1s recommended Here we present what has emerged during this prelimary phase UnicoreFTP Tests done between CINECA and FZJ by using the public Internet First figures showed sufficient results with Dataset A Many Small Files where a throughput close to the 30 of the maximum available bandwidth was achieved That is good for a public network where congestion levels are high Some difficulties have been reported for the software installation and the setup of the environment along with some problems related to PRACE 2IP RI 283493 16 23 08 2013 D10 2 Second Annual Report of WP10 reliability for long file transfers It has not been possible to run more tests on the Internet link as well as it was not possible to test the tool against the PRACE network Despite the ending of the task activities still go on and now include BSC Spain as third partner GlobusOnLine GlobusOnLine provided valuable results already during a similar test made in PRACE IIP This is not surprising since it is based on GridFTP Concerns are still related to security and in particul
47. Rate Implementation status July 2013 and export of PDF s for mailing at any point in time Enable generic export all documents related to a call to be exportable in corresponding folders files e g one folder Astrophysics containing as many as folders as proposals each containing all the documents related to this proposal application tech review scientific review 13 Keep extensive logs regarding all changes made by the Desiderata Partial users in the tool implementation 14 Provide different communication tools via email via user workspace etc between the users who have to communicate according to the existing workflow i e technical evaluator and principal investigator Desiderata Implementation in progress 15 Design and run workflows between the Coordinators of the Evaluation Process the evaluation sites and the evaluators The web based DECI tool could supoprt rule creations that would be associated with conditions and actions 1 e time reminders or enforcement establish deadlines for submission of evaluation email reminders to reviewers alerts to the evaluators of completed pending or overdue reviews Desiderata Postponed for 2014 16 Provide administrator with complete autonomous control of the tool parameters e g reopening applications needed in the administrative process changing the deadline of a review changing the discipline category of a project when the automatic
48. S Ticket Based Access Evaluation eeessssssseeeeeeeeeeee sese essen nnn nenne 66 O0 dSRODS FEDSESEVOUIOUON ee e IRR ER RI EQ S usui mex diis tos musun itcm de de EE 70 6 0 5 4RODS Performance BYalldtiOR o ae e EE E ee Re HR Wi eR UE NIS ae 72 6 6 6 iRODS Direct Access Resources Evaluation eee eene eene nnne nnne nes 77 6 6 7 RODS TD ROP CV Glug OM au toe iot t eom da mac vuU oe t Iu PE E 82 List of Figures Figure 1 Architectural diagram of the VPH use case pilot eese 11 Figure 2 File System Performance Comparision Raw Blocks serene 26 Figure 3 File System Performance Comparision Ext4 Ceph GlusterFS cccccccccccscccceeeeeeseeens 27 Figure 4 Images compressed with lossless zlib lossless jgeg and default settings 3 Figure 5 Images with jpeg compression with WAN setting 12 and 7 setting ssssss 32 Figure 6 Event Integration Screenshot 1 PRACE Training Events ccccceessscceeeeeeeeeeeeeeeeeeens 33 Figure 7 Event Integration Screenshot 2 Upcoming PATC Courses ccccsesssssceeeeeeeeeeeeeeeeeeens 34 Figure 8 Event Integration Screenshot 3 PATC Courses eeeeeeeeeeee nennen 35 igure 9 DBCT PPRETOoOLlSCteensliOls siete tte Ed a etant beatius dms ta aat aes 36 Figure 10 Graphical example for narrow and tight network links eeee
49. SU ON FUE LU vi References and Applicable Documents 5 0 25 5 2 e vrobbrkscent uva 0G Yes QCY Ver a este Va e vU Pa V dos Veo VU Vv va ao VOR na Ua vi List of Acronyms and Abbreviations sisi 5 nice veces cea VU deoa e 1s Fev eva De ra YS Oo Eve baee Ra ee Pops V ey sua Ve dae vvea Ee vii p eguiisizeria eT 1 1 eile M aa 2 2 Enhancing the Existing Infrastructure ee eee ecce eee ee eee eee eene neenon sas e tte e eese s asset eee tese e sas 2 PSP CUT TNT LUTTE 2 2 141 Central ACCOUNTING REPOSTIO Y Acosta n vae ta dtes eR EE Eco US Laas Sean tenets aa enw it Stowe 2 2 1 2 Morage ACCOUNTING arid REPOTUD usce o eie s metuo doo EN et E urn DUREE MuR Eee o bu qe din Nm Nude 3 2 0 PRACE RI Web and PRACE Event System Integration ccccsssssssccccccccccccscsceesssssssssssscees 3 DZ DEV AIG ATO E E OU UNUS 4 22 2 Toehnicul BACK OL OUNG aestu oh tu atte ot cis veiut Diouf Un un uda A Reon 4 2 2 9 ETP Pers GHG P COMIC audaces ee e t iauddacu tue tui na e bap tidie dac dues E aei ti sieudd esas 5 PLATENE a ON alae tester totae apa tedekaleates ud ediderat de fue teftis ie buie d testes felit deri iene 5 MEM cea v duili AION c 5 24 DE C NT 6 2 9 PRACE Information Portal inire etos pv aa e eo vena eden ae TE Topo aede abono aae ea ai deoa aree d Edad 7 20 LO Wd CO UIT uli 8 2 7 Collaboration with other technological oriente
50. Z PRACE gt 4 gt t SEVENTH FRAMEWORK PROGRAMME Research Infrastructures INFRA 2011 2 3 5 Second Implementation Phase of the European High Performance Computing HPC service PRACE CAPACITIES PRACE 2IP PRACE Second Implementation Phase Project Grant Agreement Number RI 283493 D10 2 Second Annual Report of WP10 Progress on Technology Scouting and Development Final Version 1 0 Author s Andreas Schott GCS MPG RZG Date 23 08 2013 D10 2 Second Annual Report of WP10 Project and Deliverable Information Sheet PRACE Project Project Web Site http www prace project eu PU 31 August 2013 30 August 2013 _ The dissemination level are indicated as follows PU Public PP Restricted to other participants including the Commission Services RE Restricted to a group specified by the consortium including the Commission Services CO Confidential only for members of the consortium including the Commission Services Document Control Sheet Title Second Annual Report of WP10 Document ID D10 2 Available at http www prace project eu Authorship Contributors Jules Wolfrat SURFsara Luigi Calori CINECA Gabriele Carteni BSC Agnes Ansari CNRS IDRIS Tom Langborg SNIC LiU Tilo Eibler GCS LRZ Giuseppe Fiameni CINECA Ilya Saverchenko GCS LRZ Miroslav Kupczyk PSNC Reviewed by Guillermo Aguirre BSC Dietmar Erwin PMO Approved by MB TB Do
51. al management of the infrastructure related to accounting service certification and monitoring Valuable input came from the collaborations with other technologically oriented project here especially the input from user communities in the pilot projects helped better understanding user needs These influenced particularly the second task This task is not fully HPC centric and more long term oriented since data management is not only of high importance for HPC users HPC generated data is also very often further processed outside HPC systems Thus many of the results achieved here do not yet bring direct improvements for the users or the infrastructure but are of importance for the further strategic decisions of PRACE concerning the handling of data in the future As already indicated in the respective sections many of the activities have potential or even needs for further investigation or development Therefore a two day hand over meeting for the important activities of WP10 to be continued in Task 6 3 of WP6 in PRACE 3IP has been held in Garching near Munich in June 2013 The relevant tasks have been identified and the planning for their continuation has been defined so the work can continue seamlessly PRACE 2IP RI 283493 32 23 08 2013 D10 2 Second Annual Report of WP10 6 Annex 6 1 PRACE Event Integration Screenshots PARTNERSHIP FOR ADVANCED COMPUTING Ing 18 Ste Home page Training and Documentation PRACE
52. ance Computing amp e Science Support Center Tera 240 1012 Bytes 8 bits also T Byte Tera 1012 Floating point operations usually in 64 bit i e DP per second also TF s Denotes the apex of a conceptual pyramid of HPC systems In this context the Supercomputing Research Infrastructure would host the Tier O systems national or topical HPC centres would constitute Tier 1 Uniform Interface to Computing Resources Grid software for seamless access to distributed resources Universal Serial Bus Virtual Imaging Platform for the Virtual Physiological Human Virtual Machine Virtual Network Computing eXtended Markup Language Hl 283493 ix 23 08 2013 D10 2 Second Annual Report of WP10 Executive Summary The major aim of Work Package 10 WP10 Advancing the Operational Infrastructure in PRACE 2IP is assessing software technologies and promoting services needed for the operation of the integrated PRACE infrastructure WP10 partly continued work of the technology task T6 3 of work package WP6 of PRACE IIP The results will be handed over to T6 3 Technical evolution of the PRACE services of WP6 Operation of the Distributed Infrastructure in PRACE 3IP This work 1s organised in three tasks covering the three goals to be achieved The first task focused on enhancing the existing infrastructure The second one concentrated on data services which go beyond the already existing ones The third task covered the remote visualization of
53. ansfers are needed Tests results for GlobusOnLine 9 and UnicoreFTP 10 are already documented in the PRACE IIP deliverable D6 3 Second annual report on the technical operation and evolution 12 This activity started by taking into account all these previous experiences The main strategy for this subtask in WP10 has been extending a perspective traditionally centered on a specific tool and considering all variables that can have an impact on transferring bulk data This allowed defining a common methodology for carrying out tests that can be applied to future and further tests of new software solutions PRACE 2IP RI 283493 13 23 08 2013 D10 2 Second Annual Report of WP10 The methodology considers factors as e Type of Dataset because moving many small files is different from moving single large files from the perspective of I O operations required e Type of Workload because tools usually have different behaviours with different size of data to be transferred e Host configuration for defining and setting up a minimum set of technical requirements for hosts involved in the communication in order to mitigate or eliminate bottleneck effects e Network capacity it is generally difficult to analyse an entire network path connecting sender and receiver but some measures along with network diagnostic tools as well as a deep understanding of the network topology can help discover the presence of any narrow link in the middle After
54. ar to users privacy because data transfer information is logged on sites that are external to PRACE Performance is good and generally between 20 and 40 of the total available bandwidth Reliability is a strong point since no failures have been registered It has been tested between CINECA and EPCC BBCP Tests of BBCP have been most extensive The largest benefit of BBCP is the possibility to install it with user privileges and asking for opening a specific port range in the firewall which can be those already open for GridFTP It has been tested between CEA CINES and EPCC on both networks Even if more tests are needed the obtained performances have been quite good and similar to GridFTP on the PRACE Network Reliability maintenance and fault tolerance have been rated good too A further investigation is required in the transfer of dataset type B Few Large Files where performance dramatically decreased in a reproducible behaviour ARC ARC has been tested inside the Swedish network of SNIC which is publicly accessible by sending data from a local site NSC to the SweStore 19 a long term storage system on a mixed network made by 1Gbps for the last mile and 10Gbps links carriers Parallel streams are not supported and only emulated for this test Another limitation 1s related to the software footprint which has been conceived for large Grid environments which do not fit the typical scenario of PRACE few large computing systems
55. as decided to implement the DECI Project Proposal Revision on a re implemented version of the same software on which the PRACE Tier 0 peer review tool is based Thus the same basic software is used for Tier O and Tier 1 calls in PRACE Concrete work started from December 2012 where a steering committee was set up for driving endeavours The new portal was then first used for running the 1 1 DECI call for proposals Out of the initial functionality requirements list see appendix 6 2 2 the following items have been addressed Functionality Rate 1 Electronic submission of project proposals Essential 2 Developers ability to programmatically redesign the forms contents and Essential their integration with the internal database 4 Provide users with complete online control of their data application form Essential user data etc and enable them to effectively view and browse their data 1 e applicants can see all their applications response letters and applications status form the portal 5 Assign different roles coordinator of the process evaluator etc and give Essential access to different functionalities 1 e evaluation assignment evaluation process views and data statistical project submission form and evaluation form according to the different privilege level 1 e evaluators can gain limited access to relevant proposals and TE This would cause different log in views for Applicants Technical amp Sc
56. ble Sign out Functionality Rate HPC PRACE Comment Europa Tier 0 Tool PPR Tool Electronic submission of project Essential Well Well proposals supported X supported Developers ability to Essential Well Well programmatically redesign the supported X supported forms contents and their integration with the internal database Web based ability form design Desiderata Well Partialy PRACE TO Planned in tool to design and change the supported supported portable kernel roadmap project submission and evaluation forms Provide users with complete Essential Well Well online control of their data supported X supported application form user data etc and enable them to effectively PRACE 2lP RI 283493 36 23 08 2013 D10 2 Second Annual Report of WP10 Functionality Rate HPC PRACE Comment Europa Tier 0 Tool PPR Tool view and browse their data i e applicants can see all their applications response letters and applications status form the portal 5 Assign different roles Essential Not Well PRACE TO Roles are coordinator of the process supported supported implemented However evaluator etc and give access to Admin UI delegation is not different functionalities i e yet available evaluation assignment evaluation process views and data statistical project submission form and evaluation form according to the different privilege
57. c gfarm ticket 505 48 FIO http freecode com projects fio 49 Latency http www ciinow com 2013 01 the truth about latenc 50 VirtualGL http www virtualgl org 51 TurboVNC http www virtualgl org Downloads Turbo VNC 52 RCM viz service user doc http www hpc cineca it content remote visualization 53 Teradici PCoIP technology http en wikipedia org wiki PCoIP 54 ParaView http www paraview org 55 Blender modeling tool http www blender org 56 MeshLab clout point and mesh editor http meshlab sourceforge net 57 Visit visualization tool https wci lInl gov codes visit 58 OpenCV Computer Vision Library http opencv org 59 UniGine rendering engine http unigine com 60 CD adapco STAR CCM CED tool http www cd adapco com products star ccm plus List of Acronyms and Abbreviations AAA Authorization Authentication Accounting ADSL Asyncronous Digital Subscriber Line AISBL Association International Sans But Lucratif legal form of the PRACE RI AMD Advanced Micro Devices API Application Programming Interface ARC Advanced Resource Connector BDP Bandwidth Delay Product BSC Barcelona Supercomputing Center Spain CEA Commissariat a l Energie Atomique represented in PRACE by GENCI France CINECA Consorzio Interuniversitario the largest Italian computing centre Italy CINES Centre Informatique National de l Enseignement Sup rieur represented in PRACE by GENCI France C
58. cal file system and the iRODS data grid The iDrop web interface 1s a suite of tools that provide individuals and groups functionality of iDrop through a web service The idrop web interface includes the idrop lite Java applet for bulk uploads and downloads and provides Java Web Start links to launch the iDrop desktop GUI It is deployed as a standard war file on any commodity Java container that supports the servlet specification such as Apache Tomcat iDrop is mainly end user orientated software although some of its functionality could be useful to administrators too Responsability SITE ROLE TASK IPB Full evaluation References 1 https code renci org gf project irodsidrop 2 http www ava com 3 http tomcat apache org Tested Components iDrop 2 0 0 Release iDrop contains web and client GUI for interacting with IRODS e iDrop Swing GUI transfer and synchronization manager e iDrop Web Interface personal cloud web interface e iDrop Lite applet transfer applet for embedding in iDrop Web significant components e idrop jnip http iren web renci org idrop release idrop jnlp 1Drop Web Start Application for iDrop desktop GUI e idrop web2 war https code renci org gf download frsrelease 157 1229 1drop web2 war pre compiled war file for deploying iDrop Web Interface e idrop lite 2 0 0 jar with dependencies jar https code renci org gf download frsrelease 157 1228 Adrop lite 2 0 0 1ar
59. categorization failed Essential Postponed 17 Communicate to the centers the info of awarded projects LDAP Project ID User Accounts etc Essential Not yet schedulled 18 Create a report of reviewers with past historical information reviews attributed and reviews in previous calls including passwords Essential Not yet schedulled 19 Create a report of all persons involved in past and present calls PIs collaborators with history call proposal ID Essential Partially implementeed 20 Guarantee a highly secure log in system highly secure password Essential Partly fulfilled 6 3 Table 6 DECI PPR tool complete requirement list Questionnaire on Big Data This survey aims to gather information on the needs for data storage and data management of users and user communities using the PRACE infrastructure with needs for orders of Terabytes or even Petabytes of data Users or Communities currently not using the Methodologically a pure on line survey is considered to not provide sufficiently valid results Thus the survey needs to be accompanied by direct interviews with users and a few representatives of user communities PRACE 2IP RI 283493 4 23 08 2013 D10 2 Second Annual Report of WP10 infrastructure should answer the questions according to their plans This information will help to develop a data strategy for PRACE and the HPC centers
60. cc ed ac uk prace GridSAFE Grid SAFE documentation http gridsafe forge nesc ac uk Documentation GridSafeDocumentation Globus GridFTP http www globus org toolkit docs latest stable gridftp 7 tgftp http work deisa eu svn general globus GridFTP tgftp current README gtransfer http www prace ri eu Data Transfer with gtransfer GlobusOnLine http www globusonline org 10 Unicore FTP http www unicore eu documentation manuals unicore6 files uftp 11 EUDAT European Data Infrastructure http www eudat eu 12 PRACE IIP deliverable D6 3 Second Annual Report on the Technical Operation and Evolution 2012 http prace ri eu IMG pdf d6 3 lip pdf 13 VPH http vip creatis insa lyon fr 8080 VPH EP 9 14 FasterData Project http fasterdata es net 15 Recommended settings for TCP variables http www frozentux net 1ipsysctl tutorial ipsysctl tutorial htmlZTCPVARIABLES 16 Enabling High Performance Data Transfer PSC http www psc edu index php networking 64 tcp tune BBCP http www slac stanford edu abh bbcp 18 ARC http wiki nordugrid org index php ARC_ middleware SweStore http snicdocs nsc liu se wiki SweStore PRACE IIP deliverable D4 3 2 Cross National Programme for Tier 1 Access Pilots http www prace ri eu IMG pdf d4 3 2 lip pdf SPIP web documentation system http www spip net InDiCo http indico software org InDiCo documentation http indico cern ch thelp html i
61. ce on NFS iniaul conditians e See 2 Direct access resource file creation through iRODS and local filesystem e nfs mounted on nfs PRACE 2IP RI 283493 81 23 08 2013 D10 2 Second Annual Report of WP10 Direct access resource at path nfs The resource created 1s named nfstestresc a dummy text file used to verify that resource 1s accessible for file operations e iput R nfstestresc testfile txt testfile txt should be created on nfs home pr11g000 path and it should have pr11g000 as the file owner ils l testfile txt prl1g000 0 nfstestresc 0 2013 07 04 11 13 amp testfile txt cd nfs home pr1ig000 Is I testfile txt rw r r root root 124 Jul 4 11 13 testfile txt The test result does not match the expected result Parameters and input data Expected result Test result The documentation 2 explains that this 1s because the irods user which put the file does not have write permissions on the filesystem location of the resource But it does not work even after giving permissions to the user and making him an owner of the nfs home pr11g000 subdirectory Conclusions The Direct Access Resource feature provides a way to have direct access to the files in a resource through the filesystem they reside on However the feature depends on IRODS users having the same accounts on the machine that hosts the filesystem and having sufficient file access rights If the access rights are lacking for a give
62. chanism for qualifying and comparing results must be available e Fach test must follow specific and defined test cases In addition to a quantitative assessment factors like reliability footprint or intrusiveness maintenance code maturity support responsiveness have been considered important as well Both the PRACE network and the public internet have been considered as target medium for testing Bandwidth Delay Product The Bandwidth Delay Product BDP has been selected as the measure to reflect a production condition BDP is a term primarily used in conjunction with the TCP to refer to the number of bytes necessary to fill a TCP path 1 e it is equal to the maximum number of simultaneous bits in transit between the transmitter and the receiver The BDP formula includes network capacity and round trip time RTT of TCP packets according to the formula BDP GByte Capacity Gbps RTT s 8 PRACE 2IP RI 283493 14 23 08 2013 D10 2 Second Annual Report of WP10 It gives a measure of the network congestion at a specific time and the ability to compare different file transfer tools under similar values for the BDP It was a must to calculate it before running a test TCP Tuning Configuring TCP parameters for data transfer hosts 1s probably both the most important and the most complicated action for improving performance in the same time Settings must take into account the available network bandwidth But too specific con
63. costis Evo ie Ro EE Foe EU aede se Eve U eH dao PED ienaa Re v Uc b deu Pa R Rd UE 32 MEME OU REST LETT 33 6 1 PRACE Event Integration Screenshots eei sevo elesee ve eese sena ge e epicurei Ven eos eas ossa ces vea deed dr aEe 33 62 DECT ru ci rS 36 6 2 1 DECI peer review tools functionnality comparison table cesses 36 6 2 2 DECI peer review tool functionalities requirements list eese nnne 40 63 Questionnaire On Big PIC SEE TL ES 41 6 4 Methodology for File Transfer EvaluatiOon e eese eee e e e eee eere eee eee eene neon essa sso tette tttnoo 44 aU M PATO IICTION 3s peices casas Hcr 44 OA PEN OE sce isa sa ie eG was oes Pa HE Settee NU 45 6 4 3 Hardware and Configuration Requirements esses essen eene eene 45 CAAM COG OLOGY cssc teat oh tu abutere ipit a E a i iS 47 Od LOS COSCS oa ettet vns delati ete Nm Huc dedisti NM im tite s 49 6 4 6 Template for testing the data transfer tool essere nnne nens 49 6 5 Performance Measurement of Remote Visualization eee eee e eee eee ee eee ee eee eee ettet ettnno 50 6 6 TRODS Evaludtoit FOLIIS ordei io eee iro e eee Pe a alere toads O AaS Ea 52 6 6 1 iRODS Workflow Objects Evaluation ccccccccccccssseessseecccccnaaessseecccceaaaasseeececseaaaasesesceessaaaaaeseeees 52 6 6 2 iRODS PAM LDAP Authentication Evaluation eee eee eee nennen nennen nes 59 6 6 3 IROD
64. cument Status Sheet Accounting 0 6 29 uy2013 Draft Data Strategy iRODS PRACE 2IP RI 283493 i 23 08 2013 D10 2 Second Annual Report of WP10 08 29 July 2013 File Transfer Technologies 09 30fuly 2013 Formatting 0 10 31 July 2013 Draft Reworking Contents 0 11 05 August 2013 Draft Reworking Contents comments Annex at the very end from annex to main document comments review comments 0 24 20 August 2013 Draft Josip for 2 2 Ilya Matteo and Jules for 2 7 Jules for 2 1 Anders for 4 3 Dietmar for 6 6 more formatting for 6 6 0 25 21 August 2013 Draft Zoltan for 3 4 Luigi for 4 deleting unused acronyms updating 2 7 with input from Giuseppe extended appendix 6 4 for clarification of 3 2 0 26 22 August 2013 Draft Additions to PIP 2 5 reworked 1RODS 3 3 reworked File Systems 3 4 0 27 23 August 2013 Draft Finalizing Introduction and Summary final review corrections and fine tuning formatting 23 August 2013 PRACE 2IP RI 283493 ii 23 08 2013 D10 2 Second Annual Report of WP10 Document Keywords PRACE HPC Research Infrastructure Disclaimer This deliverable has been prepared by the responsible Work Package of the Project in accordance with the Consortium Agreement and the Grant Agreement n RI 283493 It solely reflects the opinion of the parties to such agreements on a collective basis 1n the context of the Project and to the extent foreseen in such agreements Pl
65. d projects eee ee eee eere ee eee eene enne 9 VN M AAR ui cp 9 PRZEM Peut R IDEE 9 DOIG TANG EGC PF P Gece teense 10 VEL UICE 10 2 7 5 Pilots with user communities on data requirements eese eene eene eene nennen nnns 10 5 Evalua ne Data Services 1 ren ones vi dE brio eO eue voe bud ve eeu oe VE Ced sue Va de vob vobia A Da V Pe cue VEQVU Ru RE Ced 12 IL Data SUP ate Gy c 12 3 2 New File Transfer Technologies sivcsisiccssacsvcatsccsestscsssivessecensadsassovebscwasnssasebeebsiecastdessiecenbessssvesessniasiads 13 32 0 C OnimonMMethodolo9 V csse Dae s Ie aad AR IU e So at eet TE Rd 14 2 2 2 Dato TOS Cl dOOLS S eid oat ERU dit aaa ia cae a iM Ra ei dac E a Uus a pni 16 3 2 9 Jestbed Gefinition and preliminary TESUUS it e d ed epe ta boe e Ip ds aee eui nd 16 2 2 A Outcomes and LESSONS deant esee oe t o DESEE D teste diete ed M m eni een GO 17 33 iRODS integrated Rule Oriented Data System eee eee eee eere eee eene eee eee ee teet ettet eee e eee 18 Deel PCCM CAL CV AU AL OMS cedat estoit ce Saris ats ba Flea date est str ipsi etui eal bain Cota heli e etin dod eoo tua enn IS 355 2 TROIS W OTR SIIOD a itin dus He coded ated ea t otv idulce didit facis eta uo M dM eh fud Iden M cou eat 20 3 3 3 Involvment in the Data Strategy working group cicer 20 3 3 4 EUDAT collaboration and pilot pro
66. d requirements using these use cases In November 2012 together with EGI and EUDAT a two day workshop was organised in Amsterdam where several user communities with interoperability needs e g VPH EPOS ENES MAPPER ScalaLife VERCE DRIHM MSS have been invited to present their use cases The infrastructures in return presented their data management facilities and plans As a result of this workshop a few pilot projects have been defined each with a specific objective and with the involvement of one or more user communities and representatives of the infrastructures They all aimed at establishing a prototype to share data across the e Infrastructures and community centers for medium term storage An integration workflow driving the pilot activity 1s typically composed by the following steps e data sets are ingested and registered onto EUDAT resources this will assign a persistent identifier PID to the data e data identified by this PID are then staged onto computational resources for further processing PRACE resources are usually utilized for massive data processing while EGI ones for post processing e results produced are ingested back on EUDAT and included in community data collections PRACE 2IP RI 283493 10 23 08 2013 D10 2 Second Annual Report of WP10 At the moment two use cases with PRACE involvement are being implemented following VPH and MAPPER requirements Only mature communities have been effectively involved into the
67. d simulation jobs with multiple read and write operations at the same time running in a parallel way have been executed They consisted of several such job definitions running in at the same time utilizing different parameters each All the tests have been first run on the raw block device provided by the storage box then on a local ext4 file system on top of this block device These numbers served as baseline for the comparision with the distributed file systems Finally the entire test suite was run for each of the distributed file system selected Every run was executed 10 times Taking the average of the results of these 10 runs should have eleminated any possible jitter FIO returns about 60 different measurement values when finishing The results can be categorized as follows e For read and write operations individually o Completion latency o Submission latency o Bandwidth e For the job globally o CPU usage o I O depth distribution o I O latency distribution All of this data was collected into spreadsheets out of which more important and interesting summary information was extracted These findings will be shown in the next sub section 3 4 8 Performance Measurement Results In this sub section a selection of results is presented documenting some general conclusions which could be deduced from the measurments already Furthermore two figures visualize the general findings It was found out that even on a local file system here
68. d you estimate the overall data volume of your project 1 3 How would you estimate I O volume per a typical processing job 1 4 How is the distribution and use of scratch data job intermediate data and result data 1 5 Do you have needs for the access to structured data HDF NetCDF relational databases 1 6 Are you implementing parallel I O or a specific I O strategy 1n your simulation code pNetCDF MPI IO POSIX approach dedicated IO program 1 7 Do you have requirements for the management of the data e g the handling of metadata and the querying of files 2 Distribution of data into the PRACE infrastructure 2 Do you need to store data on multiple PRACE sites 2 2 Do you expect to reuse data stored on PRACE Access provided to Tier 0 systems through PRACE calls http www prace ri eu Call Announcements Access provided to Tier 1 systems through DECI calls http www prace ri eu DECI Projects PRACE 2IP RI 283493 42 23 08 2013 D10 2 Second Annual Report of WP10 3 Are there needs for peak storage in PRACE during a project call 3 For how long days weeks months 3 2 How much data 1s that 3 3 Where would you like to have your data stored 4 Are there needs for longer term storage in PRACE between project calls or after an allocation period 4 For how long months or years 4 2 How much data is that 4 3 Where would you like to have your data stored 5 Service Level Description 5 What are you
69. data In the first task the following services have been treated in several sub tasks The centralized accounting service Grid SAFE has been extended covering now Tier 0 and Tier 1 systems and will go into production soon Another sub task integrated the announcement and management of PRACE events directly into the PRACE web site Work on the Service Certification the PRACE Information Portal and the INCA Monitoring has been continued by further developing the products or new or additional components for them and thus led to an overall improved infrastructure The Project Proposal Revision PPR tool already in use for managing Tier 0 calls has been extended to also handle proposals of the DECI calls in Tier 1 It has already been used for the latest DECI 11 call Finally the collaboration with other technologically oriented projects has been continued especially with concrete pilot projects addressing data requirements of user communities One pilot dealt with the needs of the Virtual Imaging Platform for the Virtual Physiological Human VIPAVPH 13 the other one handled the complex data workflow of a quantum mechanical problem using local EGI and PRACE resources for the calculation as part of the MAPPER project 29 In the second task Data Services the Data Strategy group as one sub task generated a questionnaire on Big Data and proposed different recommendations for improvements in data handling of which some could be implemented
70. ded As outcome a pilot project based on the 1RODS technology was defined in the scope of the PRACE EUDAT EGI collaboration Since this pilot switched to another technical solution due to a technical issue identified by EUDAT people no further work was performed for this pilot by this task But as response to the cause of that decision a technical solution was developed and documented This document describes how PRACE and EUDAT infrastructure can communicate using 1RODS by solving the double network interface issue It 1s available on the internal documentation server and can be made available on request 3 3 5 Conclusions From the experience gained in this task the following future work can be identified e get responses from users to the Big Data Survey questionnaire to clarify the applicability of IRODS for PRACE e combine the questionnaire responses with a future PRACE data strategy e consider the existing collaborative data infrastructures to develop potential collaborations e deploy a concrete PRACE EUDAT collaboration and define the interfaces between both projects e identify in which context and in which way 1RODS could be deployed in PRACE e try to influence the IRODS developments as needed e Inthe case where relevant use cases are identified o work close enough to the users to understand their data requirements for their entire project o offer to the users a long term community and project oriented solution to their data managemen
71. ding all changes made by the users in the tool Provide different communication tools via email via user workspace etc between the users who have to communicate according to the existing workflow 1 e technical evaluator and principal investigator Desiderata Desiderata Well supported Well supported Well supported Well supported 15 Design and run workflows between the Coordinators of the Evaluation Process the evaluation sites and the evaluators The web based DECI tool could support rule creations that would be associated with conditions and actions i e time reminders or enforcement establish deadlines for submission of evaluation email reminders to reviewers alerts to the evaluators of completed Desiderata Not supported Partially supported PRACE TO Included in kernel development roadmap HPC Europa Easy to develop PRACE 2IP RI 283493 38 23 08 2013 D10 2 Second Annual Report of WP10 Functionality Rate HPC PRACE Comment Europa Tier 0 Tool PPR Tool pending or overdue reviews 16 Provide administrator with Essential Partially Partially HPC Europa Basic tools complete autonomous control of supported supported e g reopening deadline the tool parameters e g change etc are already reopening applications needed available Advanced ones in the administrative process should be better clarified chan
72. e failed with error 826000 CAT INVALID AUTHENTICATION iinit Enter your current PAM system password ils AempZone home niifitest The test result matches the expected result PRACE 2IP RI 283493 67 23 08 2013 D10 2 Second Annual Report of WP10 Detailed tests An existing PAM derived password is invalidated on client side 4 see PAM derived password does exist 2 I PP l eye S 7 nitial conditions a backup of the original irodsA is available JrodsA and a backup of original valid 1rodsA at Parameters and input data client side ils cp p irodsA 1rods ils lest procedure Expected result The ils command should work again if a backup of the original irodsA is copied back to irods directory ils Enter your current IRODS password rcAuthResponse failed with error 826000 CAT INVALID AUTHENTICATION cp p irodsA irods ils tempZone home niifitest Test result The test result matches the expected result Conclusions The PAM LDAP Authentication feature allows using already existing LDAP Authentication instead of the usual IRODS password authentication Password exchange is protected SSL is being used and subsequent to that an 1RODS generated short term two weeks password is used for other 1 commands As the PAM derived password is stored in a scrambled form in irodsA file on the client side care must be taken to protect this file to assure that an imperso
73. e project a new collaboration was initiated with the EUDAT project also involving the EGI infrastructure and new scientific communities resulting in a few pilots The following aubsections give an overview of the collaborations of which some will continue within the Task 6 3 of PRACE 3IP 2 1 MAPPER The MAPPER project Multiscale APPlications on EuRopean e infrastructures 29 aims at deploying a computational science environment for distributed multi scale computing on and across European e Infrastructures including PRACE and EGI The collaboration between the two projects started in May 2011 and was coordinated via a Task Force comprising specialists from each of the three organisations MAPPER PRACE EGI Inspire On request of MAPPER then PRACE and EGI investigated the exchange of user support requests between the EGI and PRACE helpdesks This should provide end users the ability to request support from both infrastructures with just one request Technical requirements to enable such exchange of support requests between the two helpdesks have been exchanged between the two projects but the implementation has still to be planned Some of the user communities involved in the MAPPER project applied for preparatory access to PRACE facilities These allocations give these communities the opportunity to submit jobs through MAPPER developed tools The research that the MAPPER project is pursuing has a distributed nature binding different communit
74. e very light vglsphere app included in VirtualGL and a much heavier Unigine game benchmark There have been also different TurboVNC settings of 1mage quality and compression type PRACE 2IP RI 283493 50 23 08 2013 D10 2 Second Annual Report of WP10 Tests have been performed on different network connections and with different platform clients e RVN node GPU Quadro Fx 1700 CPU core Xeon E5540 2 5 Ghz e visual node GPU Quadro Fx Tesla M2070 CPU core Xeon E5645 2 4 Ghz home urban ADSL italy remote fps lossless jpeg jpeg hi jpeg low jpeg compr Windows client zlib max quality quality 12 vglsphere on visual 190 200 3 2 2 6 12 20 vglsphere on rvn 120 140 3 22 6 15 20 UniGine on visual 30 50 0 3 1 5 4 T 19 UniGine on rvn 8 15 0 3 1 2 5 7 5 9 glxspheres 1280x960 on rvn Cineca LAN Windows client client compression server cpu client cpu o client cpu server client PLINK vncviewer fps fps lossless zlib 35 40 15 15 180 4 5 jpeg max quality LAN 95 37 3599 35 70 28 39 jpeg med quality default 90 30 40 55 60 57 58 jpeg low quality WAN 90 15 35 65 70 64 67 glxspheres 1280x960 on rvn Cineca LAN Linux client client compression server cpu client cpu client cpu k server client ssh vncviewer fps fps lossless zlib 45 50 12 8 1596 150 170 6 jpeg max quality LAN 95 20 22 9 35 40 38 39 ipeg med quality default 93 12 96 28 9 55 59 56 58 jpeg low quality WAN
75. eading writing files efficiently on large scale simulations 14 What problems do you have with data management in PRACE and in general 15 What could be done to enhance your workflow 16 Is there any important question that we have missed 6 4 Methodology for File Transfer Evaluation The following sections are a copy of the internal document Methodology for File Transfer Evaluation Ever since the formulation document is mentioned this chapter is meant The references and all other numbering are adjusted to match this deliverable s list of contents tables figures and the references 6 4 1 Introduction Objective of the sub task New file transfer technologies is to evaluate alternatives to GridFTP which is currently provided as the only core service for bulk data transfer This document aims to define a common methodology for evaluating file transfer technologies that are new for PRACE 1 e not yet officially supported There are no specific technologies specified in this document since the methodology has designed to be independent from a specific software solution The main reference for this document has been a similar work being carried out by the Energy Sciences Network ESnet operated by LLNL and funded by the US DoE j http fasterdata es net PRACE 2IP RI 283493 44 23 08 2013 D10 2 Second Annual Report of WP10 This activity is being tracked by a dedicated page in the internal PRACE Wiki 6 4 2 Defini
76. ease note that even though all participants to the Project are members of PRACE AISBL this deliverable has not been approved by the Council of PRACE AISBL and therefore does not emanate from 1t nor should it be considered to reflect PRACE AISBL s individual opinion Copyright notices 2013 PRACE Consortium Partners All rights reserved This document is a project document of the PRACE project All contents are reserved by default and may not be disclosed to third parties without the written consent of the PRACE partners except as mandated by the European Commission contract RI 283493 for reviewing and dissemination purposes All trademarks and other rights on third party products mentioned in this document are acknowledged as own by the respective holders PRACE 2IP RI 283493 iil 23 08 2013 D10 2 Second Annual Report of WP10 Table of Contents Project and Deliverable Information Sheet 111 1e esee ee eee eere eee eee e erre ee eene aeos sette tete ee eaan i Document C ontrol SD66E tas ovi QS Ve rose ET CE VI CVi dE v PAP y Cel uu ETE Q D sue Ve devo PEDE a EUREN ERE Ue i Document Status Sheet 5 uri erit deo ove baden E bVa eve IRIS Led OVE Do Teu EA EN Ere a a i DOCUMENt Ca Usi H iii dro Contents detoner E C E e O aaa iv List ot UTR 6S asioina n E V EISEOF Tables 5i ice E E oU EET ERAI e HEURE POENAE ORE ERe EUR Ma EUR FATUR VD CUIUS Va TELA SUE e TD FERE VU
77. ed by adding much more nodes to the cluster than the target number of replicas because in that case a distributed file system can stripe writes well across multiple nodes for a better overall performance This assumption should be investigated further in a later phase of the evaluation The other interesting finding was that all distributed file systems show comparable performance within a range of less than 40 deviation PRACE 2IP RI 283493 26 23 08 2013 D10 2 Second Annual Report of WP10 Ext4 versus Ceph versus GlusterFS performance 15 499 11 762 4 133 4 031 E M MN a Bee eee oe Syncronized random Syncronized random Syncronized random Syncronized random read 4KB block size read 32KB block size write 4KB block size write 32KB block size E ext4 bandwidth KB s E Ceph bandwith KB s E GlusterFS bandwith KB s Figure 3 File System Performance Comparision Ext4 Ceph GlusterFS 3 4 9 Conclusions and Plan for Further Work The use cases for possible common home directories or easily accessible shared data spaces across HPC systems in PRACE illustrate the general usefulness of distributed file systems The tests of the different software solutions concerning reliability and performance do not yet indicate clear recommendations But since the future data strategy in PRACE is still open it seems to be wise to be prepared for eventual requests for the provision of a shared storage based on a distributed file system Thus
78. eeeeeeess 45 PRACE 2IP RI 283493 V 23 08 2013 D10 2 Second Annual Report of WP10 List of Tables Table 1 DECI PPR Tool Functionality Requirements eese eren 6 Table 2 Test benches for evaluating new file transfer technologies esses 16 Table 3 IRODS testbed characteristics eessssessseseseeeeeeeneennne nnne nnnm neennehenn hene rhe nere se reser enean 18 Table 4 Matrix of test cases for file systems eu eene oce Rete tni teet ut anta det nn dad auun aut tuta gd Rue det kon Resdaus 25 Table 5 Test benches for evaluating new file transfer technologies eese 39 Table 6 DECI PPR tool complete requirement list eeeeessseeseseeeeeeeee trennen 4 Table 7 File Transfer Measures Definitions cee eeeccsecccecccescccusccneccnsccusecccseceuseccuscensecensecense 45 Table 8 File Transfer Requirements list lessen enne eene eene enne 47 Table 9 File Transfer test dases with at least 18 runs each eeeseeeeeeenen nennen 49 Table 10 Example of a filled file transfer evaluation sheet here for bbcp between CINES and CEA 50 References and Applicable Documents PRACE project web site http www prace project eu PRACE research infrastructure web site http www prace ri eu 3 PRACE IP deliverable D10 1 http www prace ri eu IMG pdf D10 1 2ip pdf Grid SAFE https prace acc ep
79. een HPC systems 3 4 1 The Use case and the Purpose of the Evaluation A common use case would be sharing user specific personal and configuration data among HPC systems This would allow users to change HPC sites between DECI calls more easily if they have to since data would be accessible from more than onel HPC site c f section 3 1 Such functionality could also reduce the need for user initiated data transfers leading to several copies of the same data on different locations Another possible use case could be to give the users a common home directory same on all sites with some preconfigured scripts and configuration files to provide them a very similar environment on every PRACE system This offers the possibilty of reviving the initial DEISA philosophy where users could maintain just one home directory shared on the HPC systems 3 4 2 Technical Requirements The previously described use cases require the file systems to offer the following features e Distributed and replicated This is needed for moving the users data from the site where it is created to another automatically e Clustered It needs to be mounted on many machines and used by a large number of users concurrently so it must have cluster functionality built in e Fault tolerant It has to handle partial downtimes when only some of the partner sites are unavailable the users should still be able to use their data on the other HPC systems e Parallel It
80. egration with iRODS and possible technical issues The iDrop Desktop GUI 1s a useful tool but it still has a lot of place for improvement The main problem is lag most likely due to network latency which doesn t happen when browsing local files Because of those delays usage can be quite difficult There are also only limited search and authentication options The iDrop Web Interface 1s fast intuitive and easy to use It also has limited search options and some authentication problems but it can include direct links to the iDrop Desktop GUI to extend its functionality To get the most out of the iDrop functionality Desktop and Web interface should be used together The lack of support for GSI limits its use in PRACE Future versions solving these deficits would make this GUI for IRODS an obvious choice for both end users and administrators 3 3 2 IRODS Workshop A workshop has been organized by GENCI IDRIS CNRS and SNIC LIU from September 26 to 28 2012 in Sweden The goal of this workshop was to gather people from a wide range of disciplines interested in data management to discuss users needs and requirements to train users and computing centers staff and to tackle the iRODS strategy and future The following speakers have been specially invited Prof Reagan Moore DICE UNC and Leesa Brieger RENCI UNC as well as Jean Yves Nief from CNRS CC IN2P3 It has been a very successul workshop with more than 30 participants including develope
81. eliability The irodsFS command has been used to access data in three ways e UNIX command line interface cp ls mv and rsync e UNIX product account apache gt apache user gt mount point owned by apache user e iRODS icommands Tests Description The tested functionality worked quite well the user mounting 1RODS home via FUSE was able to read and write data to the mounted directory from the command line with traditional unix commands in the usual way The same holds true if the access is performed by a product account such as the one running a web server PRACE 2IP RI 283493 71 23 08 2013 D10 2 Second Annual Report of WP10 A couple of limitations have been revealed e itis not possible to use icommands this is by design documented on the official web page of the feature e it was unstable with rsync stable elsewhere CINECA tested a new patch from Hao Xu see 1ROD Chat 9650 3 to resolve the second issue the problem with the use of rsync is now solved Anyway the first problem unavailability of icommands prevents the usage of this functionality for some kind of production environment where data should be writable also by iRODS Conclusions The FUSE module is working well even if it is not easy to install It is used in production environment around the world and proved to be quite useful even if not absolutely reliable it could be sometimes necessary to umount and remount the collection because sometime the
82. er has no account on the 1RODS server and if the anonymous user has been created the user should be able to authenticate to 1RODS as the anonymous user and access to the data described in the ticket as every other user Responsibility The test activity 1s carried on by CINECA References CINECA evaluated the ticket based access for irods 1 https www irods org 1index php Ticket based Access See also iticket 2 https www irods org index php iticket and 3 https groups google com d topic irod chat K3 Cbyq0C8nY discussion 4 svnz irodssvn ucsd edu AARODS clients icommands scripts Tested Components We tested the iticket functionality of i1RODS introduced first in 1RODS 3 1 We tested on an iCAT mySQL enabled iRODS 3 2 server Tested Features Ticket based access Non Tested Features Most 1RODS functionality is irrelevant for this activity Tests Phases The test consisted in creating tickets with various parameters values such as different validity in time and file size and verify their functionality Tests Environment SITE TEST ENVIRONMENT CINECA eee e 16x Intel E5530 2 4 GHz e 64GB e 1 Gb Ethernet card e Debian GNU linux 6 0 Local FS GPFS NFS 160 MB s R W iRODS 3 2 PRACE 2IP RI 283493 69 23 08 2013 D10 2 Second Annual Report of WP10 Testing Methodology We evaluated functionality easyness of use and reliability The icommands have been configured for three differe
83. es In particular the FUSE package has to be installed and configured in order to give the users the right to use it inclusion in fuse group in etc groups Once the installation is completed a user should be able to for example mount its 1RODS home collection in a mount point of its UNIX workstation and that data via traditional UNIX command line tools Responsibility The test activity is carried on by CINECA References CINECA evaluated the FUSE support for irods PRACE 2IP RI 283493 70 23 08 2013 D10 2 Second Annual Report of WP10 1 https www irods org index phphRODS FUSE See also imcoll 2 https www irods org index php imcoll and 3 https groups google com forum Z msg irod chat eL 11Q5z60t4 somesjc CQsJ Tested Components We tested the FUSE functionality of IRODS introduced first in 1RODS 1 0 We tested on an ICAT mySQL enabled 1RODS 3 2 servers with icommands of the same version Tested Features FUSE irodsFS access Non Tested Features Most IRODS functionality is irrelevant for this activity Tests Phases The tests consisted in mounting an 1IRODS home directory with irodsFS and testing its reliability with I O stress test Tests Environment SITE TEST ENVIRONMENT CINECA pu e 16x Intel E5530 2 4 GHz e 64GB e 1 Gb Ethernet card e Debian GNU linux 6 0 Local FS GPFS NFS 160 MB s R W iRODS 3 2 Testing Methodology We evaluated functionality easiness of use and r
84. est should be executed at least 3 times and the average considered as the final figure For the list see appendix 6 4 5 Template A template for the testing 1s provided in Annex 6 4 6 and adopted as a common way to collect and present results of tests 3 2 2 Data transfer tools The data transfer tests have been performed using the following four tools e UnicoreFTP a pluggable file transfer mechanisms provided by UNICORE available on many PRACE systems 10 e GlobusOnLine which comes with positive but not exhaustive feedbacks from a preliminary evaluation carried out for Tier 0 systems within PRACE IHIP 12 e BBCP a tool that is spreading among scientific communities and able to support X 509 certificates for authentication and a data parallelism without requiring a remote server 17 e ARC a Grid software developed by the NorduGrid and providing data transfer features on top of GridFTP 18 3 2 83 Testbed definition and preliminary results All tests results are tracked in a dedicated page of the PRACE Wiki and can be made available on request This activity will continue in the technology task T6 3 of PRACE 3IP After all tests have been finalized the results will be made available in a separate document Test benches involved 5 PRACE partners who started testing the 4 tools on the PRACE private network and Internet as showed in the following table UnicoreFTP FZ Germany CINECA Italy GlobusOnLine CINECA Italy E
85. estw mss Create a collection and mount that collection as a Workflow Structure Object associated with the workflow file testw mss imkdir IDRIS home pr1f02is workflow testW F imcoll m msso IDRIS home pr1f02is workflow testw mss IDRIS home pr1f02is workflow test WF Ingest a parameter file testw mpf in the WSO collection test WF iput testw mpf IDRIS home pr1f02is workflow test WF Expected result a run file is automatically created in the WSO environment that will be used for the workflow execution later on Test result ils IDRIS home pr1 f021s workflow testWF prlf 2is mssoSt demoResc 2392 2013 04 23 15 19 amp testw mpf prlf 2is mssoSt demoResc 33554412 2013 04 23 15 19 amp testw run PRACE 2IP RI 283493 55 23 08 2013 D10 2 Second Annual Report of WP10 DEED UU The test result matches the expected result Test of the workflow execution Initial conditions Parameters and input data the testw run file Launch the workflow execution using the run file Test procedure iget testw run Expected result The workflow is executed Test result Command result is gt gt gt ecrifich info File home 1RODS V ault home pr1 f021s workflow testw mss cacheDir0 testw runDir fichin 1 NBlock 128 BlkSize 5 12 Workflow ecrfich Executed Successfully at 2013 4 23 15h 14m 12s The test result matches the expected result Test of the parameter file settings Note e The
86. esults have been processed and this resulted in a report by the end of 2012 The internal report and its conclusions have been discussed in two dedicated video conferences early in the spring of 2013 The report gives an overview of the disk storage accounting policies and tools in use by sites The main conclusions and results of the report are e Less than half of the partners use disk storage accounting e The tools that are provided to users to get information on available and used storage vary from built in OS tools and specific file system tools to specific site developed scripts and open source tools e On the PRACE internal wiki a table is maintained with up to date information about the storage facilities at sites e Asanext step it is proposed to investigate in the use of a uniform tool within PRACE for the provisioning of information about actual storage utilization to users e There is no requirement from partners to further develop storage accounting facilities Task 6 3 of PRACE 3IP WP6 can use the results of this activity to further evaluate and develop facilities to display the actual storage utilization 2 2 PRACE RI Web and PRACE Event System Integration The integration of an event managing system into the regular PRACE RI Web Site came as a response to the ever increasing need to announce the PRACE events in a more efficient manner and to enhance ease of access to all the relevant information Integrating these services req
87. ethodology Ceph Ceph is well documented and the software is easy to install because the developers themselves make Debian packages from the latest versions of the software A quite complex initial configuration was required but then mounting on the client was an easy task Replication worked out of the box Technologically the replication is based on a consistent hashing algorithm and synchronized replication so the client itself is informed by the server PRACE 2IP RI 283493 23 23 08 2013 D10 2 Second Annual Report of WP10 how many replicas are needed and where to place them it does not get a write successful notice back until all the needed replicas are written to the nodes This results 1n a tradoff for the write performance but assures a high availability in a multi site setup The automatic repair function worked too The nodes synchronized data after the node with the outdated data joined back Fault tolerance was seamless the client did not even notice a server node going offline when it was writing data to it Although fulfilling every aspect of the feature validation tests the developers themselves state that parts of the file system are not yet stable enough to be used in production environments Furthermore the complexity of the configuration has to be noted GlusterFS GlusterFS 1s also well documented and has stable Debian packages available from the official repository The file system configuration has been
88. figurations can sometimes even lead to a depletion of performance especially if low speed networks are used Modern operating systems meanwhile provide an excellent auto tuning for the TCP buffers leaving a system administrator play with maximum values only Many other TCP related variables could be recommended but this 1s out of scope and similar works are suggested in 14 15 and 16 This activity focused on three important settings details can be found in appendix 6 4 3 e TCP Buffer Size Values varied depending on the available memory of the machine and the BDP see above e MTU and Jumbo Ethernet Frames Testing with a MTU of 9000 compared to the default MTU of 1500 e Disk performance Check performance of the disks subsystem with I O benchmarks like hdparm bonnie and iozone Data sets Transferring a large number of small files 1s significantly different from transferring few large files 1n terms of performance Also the directory tree affects performance significantly The methodology considers two different datasets one with many small files and the other with a few large files Details can be found in the appendix 6 4 4 Workload Testing a tool against different workloads is a good way for producing an exhaustive assessment since it simulates a wide variety of situations happening on real systems and therefore allows for detailed analysis of the behaviour in real life A huge workload can thus provide information ab
89. formance of the different VNC based services under different usage conditions see section 4 4 and the further development of the CINECA RCM 52 pilot project which aims at the simplification and the improved deployment of the TurboVNC VirtualGL 51 50 software stack and is described in more detail in section 4 3 In this context RZG has tested the CINECA RCM pilot from the applications and operations point of view RZG staff has compared the user s experience of CINECA RCM with that of a standard VirtualGL TurboVNC based solution which is operated by RZG for the MPG and which has been made available also to PRACE users for analysing their simulation data produced on RZG system in the context of DECI projects RCM was successfully tested using a Paraview application example with client software for the operating systems Ubuntu 10 OpenSUSE 11 and Windows 7 According to RZG s experience RCM addresses some of the shortcomings of standard VirtualGL TurboVNC based solutions In particular RCM provides a more convenient way to reserve and access remote visualization resources Specifically reservation of resources and tunneling through firewalls is handled more transparently and in a more user friendly way by RCM RCM also allows sharing GPU and CPU resources and thus enables virtualizing the resources 4 2 Teradici PCoIP setup at SNIC LU Teradici PCoIP technology 53 enables efficient and secure transfer of pixels including associated
90. ging the deadline of a review changing the discipline PRACE TO category of a project when the Reopening applications automatic categorization failed for the administrative process Available Changing the deadline of a review Available Changing the discipline category of a project can be implemented 17 Communicate to the centers the Essential Partially Well HPC Europa LDAP info of awarded projects LDAP supported supported compliant information can Project ID User Accounts etc be already exported but specific developments could be necessary according to LDAP schema PRACE TO Specific export process already implemented for the current schema 18 Create a report of reviewers Essential Partially Well HPC Europa Easy to with past historical information supported supported implement reviews attributed and reviews in previous calls including PRACE TO Already passwords implemented in the administrators access 19 Create a report of all persons Essential Well Well PRACE TO Already involved in past and present calls supported supported implemented PIs collaborators with history call proposal ID 20 Guarantee a highly secure log in Essential Well Partially CINES Connection in system highly secure password supported supported HTTPS stronger security of password is planed for the next version of the tool Table 5 Test benches for evaluating new file tran
91. gy towards the handling of data This was acknowledged on the management level which requested the generation of the questionnaire Further investigations and effort into the development of a profound basis for decisions are dependent on the results of the questionnaire This needs to be distributed to the users and communities and then the collected results have to be summarized properly 3 2 New File Transfer Technologies The objective of this sub task has been to carry out technical evaluations of high performance file transfer tools 1n order to identify possible and reliable alternatives to Globus GridFTP 6 which is the only trusted and supported tool in PRACE for moving large amounts of data Continuous growth in computing power is increasing the need of having a reliable data transfer service for transferring bulk data inbound and outbound of the PRACE Research Infrastructure This requirement is especially pressing for scientific data intensive applications like those belonging to earth and life sciences In the past the focus was mainly on improving performance usage and logging features of GridFTP 7 Then the tool gtransfer 8 built on top of GridFTP was developed for moving data with optimized performance by an easy to use interface Feedback received from internal surveys from DECI users and from user communities like those participating in the EUDAT project 11 pointed out that more sophisticated tools for high performance data tr
92. has only four values Hourly Weekly Daily and Every two minutes for testing purposes Set synchronizations from the list can be forced to synchronize Status bar on the main window of the interface also indicates status of synchronization 23 08 2013 D10 2 Second Annual Report of WP10 iDrop Desktop edition places an icon in the system tray with the common set of options when right clicked System tray icon iDrops Web Interface Login screen can be modified through idrop web config2 groovy file in etc idrop web Starred Files directory Default values for host port zone Starred Folders resource and authorization scheme can be set Folders shared by me and they won t be displayed on the login Folders shared with me screen Home screen offers overview of Quick upload starred files and folders shared folders and quick upload tool that uploads selection to predefined folder Login and Home screen Browse screen offers 1RODS tree view with very user friendly interface Tree context Tree context menu menu contains all the options for refreshing Add to cart creating renaming deleting cut copy pasting Bulk upload of the content and getting corresponding Tickets information Uploading can be done by Quick upload or Bulk upload for multiple selections Download is managed by the shopping cart feature that lets you store your choices and download them at any time by checking out Info view gives basic information as well as
93. hrInMb to a default value of 52 e maxNumThr The maximum number of threads to use It accepts integer value up to 16 It also accepts the word default which sets maxNumThr to a default value of 4 A value of 0 means no parallel I O This can be helpful to get around firewall issues e windowSize the tcp window size in Bytes for the parallel transfer A value of 0 or default means a default size of 1 046 576 Bytes The runs were done using a dedicated script program which performed network performance measurement before each run and executed the transfer using the 1RODS iput command several time to ensure reliability on the performance printed out Final results PRACE 2IP RI 283493 76 23 08 2013 D10 2 Second Annual Report of WP10 A 1 file of 100MB A 100MB 0 26 MB s 4 A 1 file of 100MB A 100MB A 16 3 03 MB s 1 09 MB s 0 85 MB s Bd file of 1GB B IGB Server choice _ 6 89 MB s 2 24 MB s 1 88 MB s 5 B fileof GB B IGB BQ 1 07 MB s 1 30 MB s 0 147 MB s 6 B lfileofIGB B 1GB B 4 2 06 MB s 2 24 MB s 0 579 MB s These numbers show that e CINECA transfer with 1RODS is able to provide good performance up to 70 of the peak of its opened and shared internet network link e IDRIS iRODS server is limited by the I O bottleneck on its server e NIIF iRODS server is limited by the bad network performances and the poor disk I O rate and the server resources virtual machine with reduced
94. ientific evaluators and DAAC staff 12 Create and export documents and information that should feed other systems Desiderata Or processes 1 e automatic generation and export of PDF s for mailing at any point in time Enable generic export all documents related to a call to be exportable in corresponding folders files e g one folder Astrophysics containing as many as folders as proposals each containing all the documents related to this proposal application tech review scientific review 13 Keep extensive logs regarding all changes made by the users in the tool Desiderata 14 Provide different communication tools via email via user workspace etc Desiderata between the users who have to communicate according to the existing workflow 1 e technical evaluator and principal investigator 19 Create a report of all persons involved in past and present calls PIs Essential collaborators with history call proposal ID 20 Guarantee a highly secure log in system highly secure password Essential Table 1 DECI PPR Tool Functionality Requirements Actual adaptation work from the Tier 0 tool started on April 6 2013 in close contact with WP2 s DECI program manager for the DECI 11 call which was open from May 6 to June 14 PRACE 2IP RI 283493 6 23 08 2013 D10 2 Second Annual Report of WP10 At the time this document is being edited the tool was in use for three months The adaptation
95. ies and systems Access to the PRACE e Infrastructure is required to ensure proper functionality and integration of tools and services provided by MAPPER In particular the focus is on the software middleware development and adaptation taking care of the infrastructure management rather than resource provisioning 2 2 EMI The EMI European Middleware Initiative project 1s a close collaboration of the four major European middleware providers ARC dCache gLite and UNICORE Its aim is to deliver a consolidated set of middleware components for deployment in EGI PRACE and other projects and to extend the interoperability and integration between grids and other computing PRACE 2IP RI 283493 9 23 08 2013 D10 2 Second Annual Report of WP10 infrastructures A joint work plan to implement collaboration s objectives was defined in a Memorandum of Understanding MoU which was officially signed by respective projects coordinator at the beginning of 2013 The EMI officially ended on April 2013 As part of the work plan EMI components those belonging to the UNICORE platform have been officially tested on PRACE sites CINECA FZJ and feedback in form of requirements have been sent to EMI STS Security Token Service developers 2 7 3 IGE and EGCF The Initiative for Globus in Europe IGE 30 did support the European computing infrastructures by providing a central point of contact in Europe for the development customisation provisioning
96. igation Ctrl key can be used as an option key for drag amp drop When pressed copy function is executed and move function otherwise There is also an interface button for copy move Interface button brings up an info panel for the current selection Tags and comments can be updated Metadata can be created and or deleted User can set permissions for the current selection within his rights Desktop iRODS drag amp drop Desktop to 1RODS IRODS to Desktop Option key Settings iDrop Accounts Transfers Synchronization Synchronization Synchronization mode Synchronization frequency Status bar PRACE 2IP RI 283493 85 Recursive directory and files drag amp drop from desktop to 1RODS and from 1RODS to desktop works with the same option key functionality Four tabs Show iDrop GUI on startup and Show within file transfer progress can be checked in 1Drop tab Accounts tab gives options for Default Resource Login to Another Grid and Change Password Transfers tab gives options for Transfer Management Parallel Transfer Options Buffer Options and settings for IRODS agent connection timeout with options to restore default settings List of set synchronizations is displayed with appropriate folder paths Only local to IRODS synchronization mode is operational There is placeholder for IRODS to local and bidirectional mode but they are still not implemented Drop down menu for setting Synchronization frequency
97. ine at least the Bandwidth Delay Product BDP which is calculated multiplying the capacity of the network path or the narrow link 1f any and the Round Trip delay Time RTT BDP Capacity RTT This gives a measure of the network congestion and the ability to compare different file transfer tools under similar values for the BDP Data sets Transferring a large number of small files 1s significantly different from transferring few large files in terms of performance Also the directory depth or tree affects performance significantly In general a user should be able to optimize the dataset that has to be transferred e g by using archiving compression and remote synchronization techniques Two dataset are defined to take into account these case studies e Dataset A Many Small files o Number of files gt 100 PRACE 2IP RI 283493 47 23 08 2013 D10 2 Second Annual Report of WP10 o Size of each file gt IGB o Directory tree gt 1 level e Dataset B Few Large files o Number of files lt 10 o Size of each file gt 100GB o Directory tree I level Workload There is not a specific study and or survey figuring out the average amount of data transferred across PRACE sites Independently from this lack of understanding it 1s recommended to test different size of workloads and to study how tools scale Taking into account the storage availability for this test three workloads are considered e Workload A 100
98. information on tags and metadata with editing and updating options There 1s also a Ticket feature tab Tickets are tokens to IRODS files and collections that may be shared Anyone with a ticket may access your data so you can email them or share them on social media sites There is also an option to mark files or folders as starred and these can be viewed on the appropriate link on the Home screen Profile This screen provides options for entering additional information about the user Search Search files and folders by tags Search results can be deleted or added to the Shopping cart for download Tools Tools option provides link to 1Drop desktop application Account Logout Change password and Set default resource options Shopping Cart Beside already mentioned functionality shopping cart offers options for clearing deleting and reloading added items PRACE 2IP RI 283493 86 23 08 2013 Browse screen D10 2 Second Annual Report of WP10 Discussion and Conclusions This document focused on 1Drop features and user experience and not on the setting of the testing environment or its integration with 1RODS and possible technical issues The 1iDrop desktop GUI is a useful tool but it still has much place for improvement The main problem is lag which doesn t happen when you browse local files and that is not an essential feature from the IRODS perspective Because of that usage can be quite difficult There is also an issue
99. instead of many small ones Performance peak has approached 700MB s which is 7096 of the total available bandwidth of the academic national network 3 2 4 Outcomes and lessons learnt The following key findings are the results of the preliminary tests executed during this period Testing data transfer tools require time Testing data transfer tools imply considering several factors that are not strictly related to the software solution being tested Such factors are related to a proper configuration for both sender and receiver hosts as well as considering networking and security implications Many of these factors have affected the execution of tests with a resulting delay for producing results and 1n some cases inconsistent figures due to non optimised configurations Tests are also diagnostic activities As mentioned above independently from a specific tool testing bulk data transfer between PRACE sites is a valuable and efficient way to discover possible bottlenecks on the network path and in general incorrect configurations on all actors participating in a file transfer including routers and firewalls So a good throughput is not the only valuable result also bad results can provide valuable feedback when indicating some misconfiguration GridFTP is still the leader Even though only preliminary tests have been executed what has emerged is that it was truly difficult to get better performance than GridFTP The term performance does
100. inux mtu size 10 User side requirements can be checked with online free tools like http pingtest net PRACE 2IP RI 283493 46 23 08 2013 D10 2 Second Annual Report of WP10 Requirements summary TCP Buffer sizing TCP buffer auto tuning supported Maximum Buffer Size adjusted with the BDP MTU and Jumbo Frames Network cards with MTU 9000 Disk performance I O performance better than Network performace Network Capacity for Public RIT lt 70ms Internet Packet Loss 0 Jitter lt lms Table 8 File Transfer Requirements list 6 4 4 Methodology The proposed methodology must be able to e Produce assessments in a consistent manner across different sites and different network paths e Consider production conditions and any network turbulence which might occur e Assess performance for different types of workloads and different numbers of parallel streams e Gather and record results of the evaluation for each technology by using a well defined template e Create a straightforward way to qualify and compare results e Provide well defined test cases In addition to a quantitative assessment also factors like reliability footprint or intrusiveness maintenance code maturity support should be considered and qualitatively evaluated Tests must be executed on both PRACE network and public Internet Production Conditions Before running a test a report on the network status must be taken This implies to def
101. ion NOVERSION turn off the versioning of results CLEANOUT clear the stage area after execution CHECKFORCHANGE check is the file being checked has changed since the previous execution of the workflow If the file has been changed then the workflow is executed otherwise it is not Non Tested Features Rules and micro services are used by the workflows but will not be tested as such Test Phases The test phases have been described in paragraph 6 The execution order 1s a b c Tests Environment SITE TEST ENVIRONMENT IDRIS Fedora release 17 Beety Miracle IBM System x3655 4 Proc Dual Core AMD Opteron tm Processor 2218 2 GB RAM 1Gb Ethernet card PRACE 2IP RI 283493 54 23 08 2013 D10 2 Second Annual Report of WP10 Testing Methodology The test will focus on the functional aspect So we will check whether the functionality is provided that if whether it does what is supposed to do the easiness of use the reliability and maintainability Tests Description Test of the workflow environment set up we rods user account defined as rodsadmin Initial conditions the testw mss workflow file that describes the Parameters and input data P workflow a basic parameter file testw mpf Create an iRODS collection and ingest the workflow file data type msso Test procedure imkdir IDRIS home pr1f02is workflow iput D msso file testw mss IDRIS home pr1f02is workflow t
102. ited to the 12 96 since asking for more compression resulted in unbearable artifacts The RVN UniGine tests show that there 1s no gain in optimizing image compression when the frame rate bottleneck resides on the remote GPU resources they also show how the same application can hit different limits when different resources become available applications that require most server side resources are the ones that most benefit from a remote visualization service It must also be noted that there is a non negligible load on the login node for the ssh tunnel execution in the visual queue UniGine tests this load seems to be connected to the raw volume of data transfer so directly related to the available bandwidth used which is in turn related with the image compression schema adopted and the frame rate attained Nevertheless in VNC sessions performing image transfer at full speed the load on the login node can be up to one third of that imposed on the compute node this can become an issue in case many visualization nodes are served by the same login node More details on the performance tests can be found in the appendix 6 5 5 Summary and Future Work All three tasks made good progress towards improving the infrastructure Several direct benefits for the users e g the web and portal related tasks as well as remote visualization offerings could be achieved by the first and the third task Furthermore the first task advanced the PRACE intern
103. ith various authentication systems instead of the usual 1RODS password authentication The PAM LDAP Authentication feature connects with existing PRACE LDAP Authentication The password exchange is protected with SSL and for subsequent IRODS commands an iRODS generated short term two weeks password is used This is stored encrypted in the i rodsA file on the client side which must be protected to assure that an impersonation attack 1s not possible Usage of this feature eases the work with 1RODS in the PRACE environment substantially Ticket based Access evaluation summary The ticket based authentication proved to be a very useful feature for short term data sharing easy to use and reliable No particular security concern 1s foreseen FUSE evaluation summary The FUSE 38 module works properly even if it 1s not so easy to install It 1s used in production environments around the world and proved to be quite useful even if it 1s not absolutely reliable it could be sometimes necessary to umount and remount the collections because sometimes the mount process freezed Since it provides the users with POSIX filesystem access methods to their data 1t seems to be a very attractive feature Transfer performances summary Tuning 1s always an important part of the work when talking about performance Once the network was correctly configured the 1RODS tuning part was simple and easy to do and IRODS was then able to provide good performance with
104. jects sees eene essen nnns nannten essen 21 SEE E O ST OT SEEE EEE E O A E E E A E S 2 34 He System Pech OlG Gres 2 oorr o rp ez ebbe a e a E 22 3 4 1 The Use case and the Purpose of the Evaluation esses eene eene 22 PRACE 2IP RI 283493 iv 23 08 2013 D10 2 Second Annual Report of WP10 342 Techmppal REGUIT OMENI S ove vic d HN E Ua iu o didth ta M aaa d eoe UE 22 REA SOEUR SUITE NIRE OR 22 Sad TOS EVITON EN ares det T endete a aa auos dada dvd n PURUS UN Suus Boe ut de CE Cao 23 3 4 5 Deployment and Feature Validation Testing Methodology eese 23 ROUMANIE A 23 3 4 7 Detailed Description of the Performance Measurement eese nennen 24 3 4 8 Performance Measurement Results 3s destecee vere duis desese ua tragen ee AY e Eu ete a Ce ERE MG e ved un QUERI EE MM Ua dns 235 3 49 Conclusions and Plan for Further WOFK iie reed ee o Re Sao RON SHEER ER RETE 27 4 Remote VisualiZALOni cou ooi tii naO E SN aE 21 MM Pilas Meer 27 42 Feradici PCoIP s tup at SNIC LU wsscsiccccvicssceasssed ccasacen ced cetinccasiasnseauaaedacaidbedteacascs Eee eden i eia dob dera eins 28 4 3 CINECA Remote Connection Manger iier eeezedi ee een eo to ve ae esa pero eoo vId ese DIR ee aa He Ud eacesnessidecsusendees 30 4 4 Performance evaluation of VNC based remote visualization services ecce eere eere 30 summary and Future WOEE ii
105. lt IPB jrodsipbacrs 1247 32 demoResc default Table 3 iRODS testbed characteristics Detailed characteristics of each system and the details of the technical evaluations can be found in the appendix 6 6 iRODS Evaluations consisting of seven sub sections which report about the specific set of features evaluated in parenthesis the site responsible for the evaluation is specified PRACE 2IP RI 283493 18 23 08 2013 D10 2 Second Annual Report of WP10 6 6 1 IRODS Workflow Objects Evaluation IDRIS 6 6 2 IRODS PAM LDAP Authentication Evaluation NIIF 6 6 3 IRODS Ticket Based Access Evaluation CINECA 6 6 4 IRODS FUSE Evaluation CINECA 6 6 5 IRODS Performance Evaluation CINES 6 6 6 IRODS Direct Access Resources Evaluation IPB 6 6 7 1RODS 1Drop Evaluation IPB The following paragraphs summarize each of these evaluations Workflow Objects evaluation summary The workflow objects feature provides mechanisms helping users running 1RODS workflows in an integrated environment Although the use of a parameter file allows running workflows in a different context and provides a way for the users to in principle easily interact with the iRODS environment this feature is currently difficult to use and to integrate in the users development environment Thus users will likely hesitate to use 1t in the current form PAM LDAP Authentication evaluation summary Using PAM 1RODS can be configured to interact w
106. lude a RSS feed template Fullcalendar plugin The Fullcalendar plugin creates calendars from the articles the SPIP database or the Google calendar when included in the articles or columns 2 2 4 Implementation All three integration requirements upcoming and past events and a calendar view have been successfully met and the implemented features can be accessed and used on the prace ri eu website see e g http www prace ri eu PRA CE Training Events PRACE RI website and PRACE Events System integration offers ease of access to users and direct links to PRACE Events website http events prace ri eu for the desired events both through events list and calendar view 2 3 Service Certification The main goal of the Service Certification sub task was to define and implement procedures for ensuring adequate level of quality of services within PRACE infrastructure before enabling them for users This includes verification of deployed services before offering them to the users ensuring that technical requirements are satisfied ensuring that quality standards such as operational policy are satisfied and improving the quality of offered services Within the reporting period the activity focused on finalizing the general certification procedure implementing quality checklists and test scripts for selected services and performing the certification on selected services Currently the list of services which have at least partial quality
107. management Compiling a Questionnaire In order to better identify the real needs of the users this task generated on request of the PRACE Management Board a questionnaire to cover all the data related issues identified by the users of the PRACE infrastructure This questionnaire compiled 1n March and April 2013 by data experts from several PRACE sites can be found in the appendix 6 3 It was accepted by the Management Board but there was not yet a decision on when and how to perform it Thus no answers to evaluate and base recommendations on are available yet But nevertheless during the discussions when developing the questionnaire already some possible recommendations relatively easy to implement but with a possible large impact for the users could be identified Presumably Straight Forward Improvements A uniform way of handling data inside PRACE would be beneficial This could also include the creation of project storage with bigger volume for sharing data in one site between PRACE users The timed availabilty for preserving the data within that project storage could be more flexible e g allowing for storing data between different project periods DECI calls separate data transfer nodes as front nodes to clusters could reduce the load on the login nodes speed up the data transfers and eventually also allow for some sort of background data transfers PRACE 2IP RI 283493 12 23 08 2013 D10 2 Second Annual Report of WP1
108. must be scalable by being able to add nodes and storage to the cluster 3 4 3 Search Phase Scanning through available documentation it has been searched for file system software providing the above mentioned technical requirements Based on these findings the following file systems have been selected for further evaluation e Ceph QFS Gfarm GlusterFS Lustre Coda XtreemFS QFS 46 and Lustre 45 documented that their architecture has a single metadata server which cannot be clustered or replicated so they do not meet the fault tolerance requirement PRACE 2IP RI 283493 22 23 08 2013 D10 2 Second Annual Report of WP10 and have therefore been excluded The others appeared to be suitable according to their documentation 40 41 42 43 and 44 and have been selected for further evaluation 3 4 4 Test Environment Since it is difficult if not impossible to create a test environment on a HPC production system a non HPC test environment with a close by storage system has been established at NIIF This should have reduced most environment related effects like lags based on network latency NIIF operates a Fujitsu Systems DX90 S2 Storage box directly connected to the facilitated computing hosts These hosts contain Intel 17 processors and have 12G memory KVM 1s used for the virtualization of the respective cloud infrastructure The virtual servers running the latest Debian Squeeze have been created with eight dedicated s
109. n user the files on the system will be owned by the root user Since the IRODS server must run as root for the direct access resources to work the need for this feature should be carefully weighed against possible security concerns 6 6 7 IRODS IDROP evaluation Introduction This document is within the scope of the IRODS sub task in T10 2 Evaluating data services It provides a homogeneous way to evaluate features and softwares for a better coherency within the 1RODS working group This document evaluates the 1Drop Graphical User Interface and its functionality within the 1RODS data grid Purpose iDrop is a user friendly desktop GUI that manages data movement and synchronization It provides a graphical view of an 1RODS data grid supporting drag and drop transfers between iRODS and the local file system as well as data movement within an iRODS grid iDrop uses the Jargon core client libraries to establish a direct connection to IRODS via the iRODS XML client protocol This provides for efficient data transfer including the ability to use the parallel data transfer algorithm Using the iDrop GUI users can PRACE 2IP RI 283493 32 23 08 2013 D10 2 Second Annual Report of WP10 Put files to IRODS from the local file system Get files from 1RODS to the local file system Create delete rename 1RODS files Copy and move files in IRODS Replicate 1RODS files Manage the automated synchronization of directories between the lo
110. nation attack cannot be made Although there is a way iadmin rpp for the 1RODS admin to remove the PAM derived password for a user it is still advisable for clients to remove the irodsA file e g by issuing 1exit full when it is not needed and to keep away from making backup copies of it As a side note PAM can be configured to interact with various authentication systems so iRODS could be integrated with those as well However integration with other systems is outside of the scope of this evaluation 6 6 3 iRODS Ticket Based Access Evaluation Introduction The goal of this document is to evaluate the Ticket Based Authentication functionality feature This is a new feature first introduced in IRODS 3 1 Purpose The tested feature should allow end users to share data with other people for a limited amount of time or for a limited number of times or till the permission is revoked After receiving a ticket 1 e a string it should be possible to exploit it as an authenticated user as well as an anonymous user if such users exists on the server PRACE 2IP RI 283493 68 23 08 2013 D10 2 Second Annual Report of WP10 For example it is possible to create a ticket in order to give read access to a given IRODS collection two times for the following two days The ticket could after its creation be sent to a user who in the next a couple of days could use it two times to read the content of the given 1RODS collection It the us
111. ndex html OGF GLUE 2 0 Specification http www ogf org documents GFD 147 pdf REST http en wikipedia org wiki Representational State Transfer 26 JSR 268 http en wikipedia org wiki Java Portlet Specification INCA Monitoring http inca sdsc edu DART http www prace project eu Accounting Report Tool MAPPER http www mapper project eu IGE project http www ige project eu Globus Toolkit http www globus org toolkit EGCF http www egct eu PRACE 2IP RI 283493 vi 23 08 2013 D10 2 Second Annual Report of WP10 EGI European Grid Infrastructure http www egi eu EUDAT http www eudat eu iRODS workshop http www prace ri eu 1IRODS workshop iRODS workshop Agenda https www irods org index phpAhRODS User Group Meeting 2013 iRODS User M eeting Agenda 37 PAM http en wikipedia org wiki Pluggable Authentication Modules 38 FUSE http en wikipedia org wiki Filesystem in Userspace 39 1IRODS FUSE module https www irods org index php AiRODS FUSE 40 Gluster FS http www gluster org docs 41 Coda FS http www coda cs cmu edu doc html index html 42 Ceph FS http ceph com docs 43 Gfarm FS http datafarm apgrid org document 44 Xtreem FS http www xtreemfs org all features php 45 Lustre FS http wiki lustre org index php Lustre Documentation 46 SAM FS https wikis oracle com display SAMOFS home 47 Bug in Gfarm http sourceforge net apps tra
112. nditions Expected result With a valid PAM derived password ils should succeed ils Test It xm AempZone home niifitest The test result matches the expected result Detailed tests A valid PAM derived password does exist 2 PRACE 2IP RI 283493 63 23 08 2013 D10 2 Second Annual Report of WP10 Initial conditions see PAM derived password does exist 1 ShowPPtimes will be used username that is niifitest zone name that 1s tempZone PAM asks for the password N niifitest A admin Parameters and input data Test procedure A iadmin rpp niifitest iquest sql ShowPPtimes niifitest tempZone N ils cp p irods irodsA oldauth iinit diff q oldauth irods irodsA ils A iquest sql ShowPPtimes niifitest tempZone Expected result If an administrator issues an admin rpp remove PAM derived Password command for the user then icommands should not work until re authentication iinit The ils command should prompt for IRODS password It should fail because there 1s no IRODS password for user The iinit command should prompt for PAM password and succeed if the password is valid On success a new PAM derived password is generated On client side the new irodsA differs from the backup The ils command should work now iadmin rpp niifitest iquest sql ShowPPtimes niifitest tempZone No rows found ils rcAuthResponse failed with error 827000 CAT INVALID USER cp p i
113. nnual Report of WP10 Conclusions In March 2013 a one day workshop was organised in London for a first evaluation of the progress and to plan the next steps For PRACE the most important results of the pilots are e GridFTP is the data transfer protocol that can be used on all three infrastructures e The use of Persistent Identifiers PID for the registration and access of data from PRACE on the EUDAT infrastructure was implemented The pilots have formally ended August 2013 but the collaboration among the three infrastructures will continue when needed 3 Evaluating Data Services 3 1 Data Strategy Initial Situation The ever increasing amount of data produced in projects computed on PRACE resources makes it more and more difficult to transfer the data in and out of the PRACE systems The increase of the data volume is growing drastically faster than the available data transfer speed Therefore a more flexible data strategy in PRACE is very important Reasons for the need for a Data Strategy in PRACE PRACE users spend more and more time on storage issues Volume grows rapidly Moving data between storage systems can no longer be handled by users interactively Open data access demands No uniform handling of data inside PRACE Including other partners handling data EUDAT National storage etc Central handling of distributed storage can better utilize available resources PRACE users should focus on scientific work not data
114. nnual Report of WP10 in a new version 2 4 of the Service Catalogue which is submitted for acceptance to the PRACE Hosting Members by WP6 of PRACE 3IP In October 2012 the PRACE Security Forum completed a risk review of the new service with as result that there was no objection to run this service User documentation also was produced and reviewed and will be published once the service 1s going into full production Pre production tests have been prepared and run The results have been used to correct errors for some sites An additional feature was added which enables partners to start automatically a new update for the last months This can be needed if local data was updated e g because usage was reimbursed for jobs The development of a certification procedure for Grid SAFE was started 1n the sub task Service Certification but has to be completed yet All partners that have a local PRACE accounting service can now export data to the central service At the moment 14 partners export their data 2 1 2 Storage Accounting and Reporting The objective of this activity was to analyse the need and possibilities of storage accounting and disk usage information for users A survey was prepared in the first project period to collect information from all PRACE partners and AISBL on this subject The survey see appendix in deliverable D10 1 3 was issued to all partners sites in October 2012 of which twenty partners have responded The r
115. ns and comments will be given if needed iDrops Desktop Swing GUI Login screen is simple and functional with all the necessary fields and responds to all Cancel login stated situations 1n a proper manner and with Bad host uid password proper notifications Launching second instance Tree Standard desktop functionality Path size and last modification time displayed when cursor Browsing and selecting local files and is put over a file or folder in the local tree folders and getting relevant info Creating renaming deleting local folders Recursive deleting Rods Tree Same functionality as with the Local Tree but with notable lag Uploads and downloads can PRACE 2IP RI 283493 84 23 08 2013 D10 2 Browsing and selecting files and folders and getting relevant info Creating renaming deleting local folders Upload and download of files and folders Progress bar Refresh button Tree root Copy amp Move Option key for drag amp drop panel Tags and comments Metadata Permissions Second Annual Report of WP10 be performed via drag amp drop or interface buttons Progress bar indicates transfers When uploading and downloading empty folders status bar doesn t indicate progress but remains at 096 Refresh button exists but it is usually not necessary as 1Rods tree refreshes by itself Refresh function maintains expansion of the file tree There is a drop down menu for setting the tree root for easier nav
116. ns such as ParaView 54 Blender 55 Visit 57 OpenCV 58 MeshLab 56 and others It supports pre compiled codes as the UniGine 59 graphics engine test as well as pre compiled ParaView deployment but there have been some issues with StarCCM 60 visualization code 4 4 Performance evaluation of VNC based remote visualization services In all visualization applications the overall satisfaction of the user interacting with the system is the most relevant criteria for the evaluation of the system Therefore the most important parameters for the evaluation are those effectively perceived by the user e the effective frame rate at client side e the overall latency of the system e the visual quality of the image stream It is important to underline that these parameter must be measured taking into account all the components that compose the client server system Server side hardware platform CPU GPU Application code OpenGL interposition layer VirtualGL VNC image compression TurboVNC server Network transport depends heavily on network bandwidth VNC client platform for image decompression and stream rendering PRACE 2IP RI 283493 30 23 08 2013 D10 2 Second Annual Report of WP10 It was decided to concentrate on the frame rate parameter as the other two even if very important in determining the overall user satisfaction are much harder to estimate in a quantitative way Almost all the VNC clients use aggressive loss
117. nsfer others less contributing are audio USB bridging and to an almost negligible extent system management Network latency up to 150ms are supported and responsiveness typically gets sluggish around 40 60ms This is however subjective and session dependent 4 3 CINECA Remote Connection Manger The Remote Connection Manager CINECA pilot project has already been described in an annex included in the previous deliverable D10 1 3 The system is available since almost one year on the CINECA PLX cluster nodes and has been recently enhanced to support new graphics nodes different access modes and has also been used to support non accelerated VNC sessions on front end nodes of CINECA Blue Gene Q Tier 0 machine The client part consists of a single executable that wraps the TurboVNC client and the python code dealing with ssh tunneling needed to support visualization services installed in compute nodes that are not directly accessible The client supports re connection to open sessions and PAM authentication It does not handle session sharing or VNC password The client is able to auto update when a new version is available The server side currently supports session book keeping and has support for PBS PLX cluster LoadLeveler Fermi BG Q as well as direct ssh access The code is available from the web under https hpc forge cineca it svn RemoteGraph trunk The service has been tested with different open source visualization applicatio
118. nt users e the owner of the file password authenticated user e asecond different password authenticated user e an anonymous user Tests Description The tested functionality iticket iget iput worked as expected for files and directories resulted easy to use and reliable but a bug has been found with the subdirectories transfer see 3 IROD Chat 9990 the recursive transfer of subdirectory does not work This was preventing the usage of this functionality for a production environment CINECA developed a set of wrapper around the icommands bash scripts to produce a workaround This workaround is now on 1RODS SVN 4 Conclusions The ticket based authentication proved to be a very useful feature easy to use and reliable Even if it had some initial problem it is now possible to use it as expected No particular security concern is foreseen 6 6 4 IRODS FUSE Evaluation Introduction The goal of this document is to evaluate the FUSE feature FUSE was introduced in 1RODS 1 0 This feature works for the IRODS client it adds the irodsF s command to the icommands Purpose The tested feature should allow end users to mount their 1RODS home directory on each machine where the feature is installed In order to enable the functionality it is necessary to recompile the icommands no package 1s available yes after having enabled the functionality in the 1RODS configuration file and having installed the required librari
119. ntly two different appliances are used an EVGA PDO2 zero client and a Samsung 24 monitor with integrated PCoIP client 1 e the monitor connects directly to the Ethernet socket The current setup is point to point and serves power users at the campus with a high performance secure remote visualization mechanism No long distance WAN tests have been possible to perform Main application area is post processing of large CAE data sets using software such as Abacus CAE and Paraview From a user experience it 1s equal to using a local workstation with respect to authentication and usage but of course much more powerful since the system is an integrated part of the computational cluster Then main operating system in use 1s Centos but one of the visualization nodes has been running MS Windows as part of the test An important benefit that distinguishes this setup from software based solutions is the remote visualization subsystems independence from the host computer as described above in further detail No specific software or drivers need to be loaded and hence there is nothing that might conflict with the operating system or end user applications Furthermore the solution puts no additional load on the host such as CPU cycles needed for image compression host to graphics bandwidth for image readback etc This allows the application to run at full speed as if displayed to a local monitor Achieved remote image quality is only determined by a
120. o Also the ease of use reliability maintainability and security will be evaluated PRACE 2IP RI 283493 60 23 08 2013 D10 2 Second Annual Report of WP10 Tests Description Basic PAM test niifitest user account in 1RODS defined as rodsuser niifitest user in LDAP niifitest has password in LDAP only Parametersandinputdata username that 1s niifitest d PAM asks for the password Test procedure cd IRODS_HOMF server bin E PamAuthCheck niifitest Initial conditions Expected result PamAuthCheck waits for password Then it tells whether the user authenticated successfully or not PamAuthCheck niifitest Test result wrongpassword Not Authenticated PamAuthCheck niifitest KK K K K K K K Authenticated The test result matches the expected result Detailed tests PAM derived password does not exist 1 see Basic PAM test PAM derived password does not exist additionally a custom iCAT query created by admin user will be used to check expiry timestamp 1 e lifetime as well as creation modification timestamps of PAM derived passwords on server side iadmin asq select pass expiry ts R USER PASSWORD create ts R USER PASSWORD modify ts from R USER PASSWORD R USER MAIN where user name and zone name and R USER MAIN user id Initial conditions R USER PASSWORD user id ShowPPtimes username that is niifitest Parameters and input data n zone name that is tempZone N
121. o be recompiled before testing It can be downloaded at 7 PRACE 2IP RI 283493 53 23 08 2013 D10 2 Second Annual Report of WP10 iRODS rule files will be used for testing the workflow execution and random data test files provided at 8 will be used for testing the stage in out and copyout functionalities Tested Features The workflow objects evaluation will focus on 3 major points a The set up of the IRODS workflow environment The tests will consist of evaluating how the 1RODS workflow environment is built before the workflow execution The set of commands to build the environment will be run checking for the real impact b The workflow execution The test will consist of executing a workflow and checking how the workflow environment is modified This test will be performed with a basic parameter file c The various parameters file settings The tests of the various parameters described in the workflow parameter file will be evaluated NPARAM describe a input parameter FILEPARAM identify files that are used as input parameters INPARAM It is needed to stage back outputs STAGEAREA identify the stage area where the workflow execution is performed STAGEIN stage in files from anywhere in iRODS to the stage area STAGEOUT move files from the stage area to the IRODS WSO COPYOUT leave a copy in the stage area and make a copy in the IiRODS WSO useful 1f it is needed for subsequent workflow execut
122. ocese nini PRACE Training Porial Eunmmureer cat HPE PRACE Advanced Training Centres er Message Passing Programming with MPI PRACE Seasonal O EPEC Schools PRACEWhiepapers PATO Laoaureez Lipcuam ng Parier ELS Past Upcoming PATC Courses PRACE TrB nsm event Best Practice Guides PRACE Ler Deere i eon Canierences ane meetings Past PATC Courses PRACE industrial Sears PRACE Outreach Events Home page Coninct Site Magn Steiessics vices WKE HPC Reinied Events c Ejes E3 resina and nocumerisian E PRACE Taming events E FATE couse o3 Boe 7 Media Sie crenied with SPIP 2 1 234 AHUNTSIC Press Maier a C Copyright PRACE AISEL 2010 all ngis reserved eed Hosted by CINES New wc Putiicatons i Presenimbansabevent Figure 8 Event Integration Screenshot 3 PATC Courses PRACE 2IP RI 283493 35 Europe place na Global Race PRACE Newsletters WATCH LIVE HOW TRACE SUMMER SGHOOL 2013 Puy 0 AUTUMN SCHOOL 2013 23 27 September jubljana Slovenia PRACE Hews PRACE Summer Schoal Ostrava Czech Republic 18 June 2013 PRACE Educntez THT 5 Researchers 14 June 2013 Welcome to the PRACE booth ai150713 Exniimson 14 June 2013 DECL11 Cal desing deadline exiend bp 14th June 173001 7 June 2013 The PRACE 5cientrhic Canseco wif oor ss stories and staiegies in European HPC 8 June 2013 PRACE related
123. ollowing information providers e Network Monitoring e INCA e LDAP The implementation covered the mechanisms for gathering of the information originating from different PRACE sources processing and finally putting them into the database For the better understanding of the collected data and the debugging process PSNC created a simple test web portal It 1s accessible for PRACE users with their certificate under the following address https dmon prace fe srv lrz de More information providers will be combined into the future production PIP portal In the test portal one can view currently e PRACE Link Availability e PRACE Link Latency e System Information currently software versions and service availability Work achievements and details on all ongoing activities of the task are thoroughly documented in PRACE Wiki 2 6 INCA Monitoring Within PRACE the Inca Monitoring based on the Inca software originally developed by SDSC 27 1s constantly updated by WP6 to match the current state of the PRACE infrastructure Besides this within WP10 work is conducted to further improve the user interface offered by Inca monitoring A complete coverage of all services and tools employed within PRACE should be achieved as the final goal Therefore existing Inca reporters originating from the DEISA project needed to be adapted to the PRACE infrastructure Namely the Inca reporter for the version of the FFTW library was re implemented
124. onment so that users can hesitate to use it 6 6 2 IRODS PAM LDAP Authentication Evaluation Introduction This document is within the scope of the iRODS sub task in WP10 Task T10 2 Evaluating data services It provides a homogeneous way to evaluate features and software for a better coherency within the 1RODS working group This document focuses on a new feature in iRODS 3 2 PAM LDAP Authentication Purpose PAM provides a mechanism for applications to detach the task of authentication from the application itself Most of the time organizations already have their AA infrastructure and it is a common case where LDAP is included in such a setup A PAM LDAP module therefore could be used to authenticate existing LDAP users In the iRODS context users may authenticate themselves using their LDAP password instead of having a separate password in iRODS For the sake of security the password exchange is protected SSL is being used and then an iRODS generated password is used which is valid for two weeks by default These so called PAM derived Passwords may be removed by an administrator for specific users Evaluation will focus on the setup the ease of use and security aspects of the feature User base for this feature includes end users as well as administrators Responsability SITE ROLE TASK NIIFI Full evaluation References 1 https www irods org index php Release Notes 3 2 2 https www irods org index php P
125. ould be copied in bin The file owner becomes the IRODS admin The copy is performed properly for a file in the WSO or anywhere in the 1RODS environment The test result doesn t match the expected result Initial conditions NPARAM FILEPARAM parameters set Parameters and input data NPARAM Arg fichin 1 FILEPARAM Arg ils l IDRIS home pr11f021s workflow testW F testw runDir prlf 2is mssoSt demoResc 741 2013 04 23 15 22 amp stdout prlf 2is mssoSt demoResc 262152 2013 04 23 15 22 amp fichin 1 The test result matches the expected result Note If the parameter FILEPARAM is not set the fichin 1 file will remain in the stagearea bin and will not be copied back in the WSO The parameter STAGEOUT fichin 1 has no impact effect in this case Initial conditions STAGEOUT parameter set Parameters and input data the file to stage out from the stagearea to the WSO Test result The file is staged out as expected PRACE 2IP RI 283493 37 23 08 2013 Test result D10 2 Second Annual Report of WP10 The test result matches the expected result Note There is currently only one stagearea available to the users bin We found here a security issue as any file can be staged out from this directory by any user all files are owned by the IRODS administrator Please note that no error message appears if the file that has to be staged out doesn t exist for any reason so cannot be staged ou
126. out stability and reliability of a software solution as well as features like checkpoint and restart of a file transfer after a failure The methodology defines three different workloads ranging from 100GB to I TB For details again see appendix 6 4 4 Parallel Streams Only tools supporting or emulating data transfer parallelism have to be considered Choosing the number of parallel streams is strictly related with the capacity of memory on both endpoints A wrong number e g an extremely high one chosen with the idea that more parallel transfers will improve performance can produce a negative consequence with a significant decrease of the data transfer rate Thus the methodology considers three different values for the number of parallel stream 4 8 and 16 Qualitative Factors Qualitative factors which are not strictly related to the data transfer rate are also able to provide important information for rating a data transfer tool like reliability or community acceptance for more see appendix 6 4 4 The methodology makes use of a 5 level ranking mechanism from 1 really bad to 5 really good along with a short comment or feedback provided by the tester that motivates the evaluation PRACE 2IP RI 283493 15 23 08 2013 D10 2 Second Annual Report of WP10 Test cases As a result of combining all different setups for datasets workloads and parallel streams the total number of runs for each test 1s equal to 18 The same t
127. peg med quality default 1 396 5 3 2 40 50 1 2 jpeg low quality WAN 299 796 294 799 39 50 2 3 ipeg ultra low quality 12 1 396 10 2 7 40 60 5 6 NICE DCV setup UniGine 1024x768 on visual low speed ADSL Windows client client compression client cpu vncviewer server fps client fps default settings low quality 1096 40 70 UniGine 1024x768 on visual Cineca LAN Windows client client compression client cpu vncviewer server fps client fps default settings adaptive 28 40 70 6 6 iRODS Evaluation Forms The following sub sections details the evaluations results of the different IRODS functions The seven sections are originally seven individual documents The references therein refer to the References section of the respective document sub section 6 6 1 iRODS Workflow Objects Evaluation Introduction The goal of this document is to evaluate the workflow objects feature Workflow objects are a new feature introduced in IRODS 3 2 Purpose The common definition of a workflow 1s that it allows to chain and control tasks in order to perform a complex processing It consists of a sequence of steps which execute a series of computation or data manipulation Each operation in the step may use data provided as input to the workflow or data created in a previous steps and may produce output data In the 1RODS context the operations in a workflow objects can be mapped to micro services so that a com
128. pilots for basically two reasons a to limit the effort into few months of work so to only focus on achievable and operative goals b to work with communities which already experimented the integration of different services and got stucked really close to complete their plan VIPAVPH The goal of this project is to offer imaging scientists a convenient mechanism to access computational and data resources ensuring the sustainability of image simulation workflows beyond a particular computing infrastructure and workflow technology This is realized providing an interoperability layer between the Virtual Imaging Platform and the VPH toolkit 13 Multi modality medical image simulators MRI US CT and PET are described as workflows using the MOTEUR technology which is able to access several infrastructure services seamlessly The pilot worked to achieve two main goals a identify which sites being part of involved infrastructures could commit their resources for the community b develop a mechanism to easily ship data across the sites To address the second goal the GridFTP protocol was selected and client adapted to handle data transfer using the EUDAT PID Currently the participating sites have been confirmed EPCC will make available its PRACE resources and data transfer performance across the sites are under evaluation MAPPER The goal of the project 1s to develop computational strategies software and services for distributed multiscale
129. plete workflow can be seen as an 1RODS rule that will be executed at the server side The implementation of a workflow is depicted as an 1RODS active object where both a workflow file and a set of related parameter files are ingested into IRODS A WSO Workflow Structured Object is associated to each workflow file for providing an 1RODS PRACE 2IP RI 283493 22 23 08 2013 D10 2 Second Annual Report of WP10 collection type environment for running the workflow The WSO will gather parameters files needed to run the workflow as well as input files needed for the workflow execution A set of run directories will be attached to this structure that houses the results of executions The WSO is created as a mount point in the 1RODS logical collection hierarchy A parameter file contains information needed for executing the workflow as well as information about files that need to be staged in before the execution or staged out after the execution When a parameter file 1s ingested into a WSO a run file 1s automatically created which will be used to execute the parameter file with the associated workflow When a workflow execution occurs a run directory is created automatically for storing the results of this run Unlike other software the 1RODS implementation of workflows doesn t provide a graphical interface used to model the various steps of the workflow Rather it is based on the standard iRODS rules and micro services Additionally it manages au
130. r team members computing centers staff and end users The 1RODS workshop agenda can be found at 35 The major data management needs and requirements that have been discussed during the workshop are e sharing preservation data transfer replication metadata management publication data mining workflow storage data volume EUDAT collaboration 3 3 3 Involvment in the Data Strategy working group several partners involved in the 1RODS task have been also involved in the Data Strategy working group in order to generate a data survey questionnaire aimed to better understand the user needs and requirements regarding the data management within the PRACE project Some of these partners have been working for PRACE only most of them have been working both PRACE 2IP RI 283493 20 23 08 2013 D10 2 Second Annual Report of WP10 for PRACE and EUDAT projects Several topics in this questionnaire such as the data sharing may highlight the 1RODS benefits for users compared to a simple data transfert tool User answers to this questionnaire as well as finding iRODS pilot projects will define if it is worth deploying 1RODS within PRACE as general service This questionnaire c f section 3 1 can be found in the annex 6 3 When the questionnaire will be submitted to the end users is not yet decided 3 3 4 EUDAT collaboration and pilot projects The data management workshop organized by SURFsara from November 26 to 27 2012 in Amsterdam was atten
131. r reliability requirements safe storage versus can be recovered easily by other means 5 2 What are your availability requirements always online access versus archived data 5 3 Is versioning for the data required 5 4 Are there any special privacy restrictions required on the data Read only Read Write access rights depending on user categories strongly enforced access rights data encryption Data Movement 6 How much data do you need to import and export from and to PRACE for your calculations volume frequency 6 1 Projects in the coming years 6 2 Jobs related to these projects 7 Where does your data come from and where do you send your data from PRACE 7 1 Within the PRACE infrastructure 7 2 From to your own site or scientific large scale equipment telescope sequencer accelerator network of sensors 7 3 From to another HPC or data infrastructure Eudat EGI national regional centers XSEDE 7 4 Do you know how fast your internet connection to the PRACE infrastructure 1s 8 Tools protocols required supported for data transfers 1f this sounds too technical to you just don t answer 8 1 Which tools are supported at your site infrastructure 8 2 Which tools are you familiar with and using 8 3 Which other tools are you interested in or want to use Data Sharing 9 Ifyou share or need to share your data with other users groups or communities 9 Is data to be shared among PRACE systems 9
132. rate sub sub tasks which are described in the following paragraphs The goal of the first one was to work on detailed technical evaluations of the current 1RODS release 3 2 which was made available in September 2012 In this context the IRODS User Group Meeting 2013 from February 28 to March 1 in Garching was attended The agenda and presentations can be found here 36 The second one focused on the information dissemination workshop on a large contribution to the data strategy working group and on pushing external collaborations such as with the EUDAT project In that workshop the contacts to the developers have been strenghtened and the planning for the development could be influenced Futhermore users have been instructed on the possible use of these tools for their future data management Another concern of the 1RODS sub task was to provide a methodology during the technical evaluation phase For this reason a Feature and Software Evaluation Template has been defined which was used to provide a homogenous way to lead the evaluation process This template 1s used for evaluations found in the appendix 6 6 iRODS Evaluation Forms 3 3 1 Technical evaluations The technical evaluations have been performed on a testbed set up between five sites while the details of the testbed can be seen in the Table 3 Address n version Ee eect lun oe 32 Poena n 1247 BR NIIF jirodsOl niifhu 1248 NIF 32 niifData defau
133. reafter a new test is performed to evaluate the setting of a given parameter e A new parameter file is created and ingested into 1RODS thus producing each time a new run file Indeed testing the different parameter settings on the same file introduces some edge effect that we didn t want to face e For each setting the test procedure and the expected results are described in the same way so we will detail only the parameters and input data the initial conditions and the test result Initial conditions STAGEAREA parameter set Parameters and input data the X run file Test procedure Launch the workflow execution using the run file iget X run Expected result The workflow is executed using the parameters describe in the parameter file Test result In the current release the STAGEAREA parameter only refers to the bin directory of the IRODS server Changing PRACE 2IP RI 283493 56 23 08 2013 D10 2 Second Annual Report of WP10 this value introduces a wrong behaviour The test result doesn t match the expected result Note The stagearea 1s defined on the local machine Initial conditions STAGEIN parameter set Parameters and input data he file to stage in which can be in the WSO environment or anywhere in iRODS Test result The stage in action on a file copies the file in the bin cmd directory of the IRODS server This directory 1s not the STAGEAREA even fixed to bin for now so sh
134. resources cpu memory network and disk I O e Threads specification at runtime as a parameter to the iput command do not give better results than default IRODS settings because of the good tuning of the 1RODS server configuration Conclusions The tuning is always an important part of the work when talking about performance Once the network was correctly tuned the 1RODS tuning part was simple and easy to do and iRODS was then able to provide good performance without any runtime setting iRODS appears here as a simple tool for transferring files allowing good performance Performance tests should be continued including additional test cases on the PRACE high performance 10 Gb s dedicated network to show the full 1RODS performance capacity Despite the testbed caracteristics which were not fitting with the performance goals of this work 1RODS has shown that it was able to provide good performances It stays a real and good challenger to the standard transfer tool gridFTP offering by the way much more powerful functionalities than only transfer with a simple filesystem like command approach 6 6 6 IRODS Direct Access Resources Evaluation Introduction This document is within the scope of the IRODS sub task in T10 2 Evaluating data services It provides a homogeneous way to evaluate features and software for a better coherency within the IiRODS working group This document evaluates the Direct Access Resources feature of IRODS 3 2
135. result iadmin Ir demoResc testresc The test result matches the expected result Test procedure Test result Direct access resource file creation and access through iRODS and through local filesystem e testresc direct access resource created see 11 1 Direct access resource creation e user prlig000 should be created on the unix domain in which irods instance is created and in the irods service itself e Direct access resource at path opt rodshare Parameters and input e adummy text file used to verify that resource 1s data accessible for file operations Test procedure e iinit enter pr11g000 s password e iput R testresc testfile txt Expected result The user which created the file and put it into 1rods should also be the owner of the file in the local filesystem on the resource location ils IPB home pr11g000 testfile txt cd opt rodshare home pr1ig000 Is 1 File testfile txt Size 124 Blocks 8 IO Block 4096 regular file Device 803h 2051d Inode 2501436 Links 1 Access 0664 rw rw r Uid 501 prlig000 Gid 501 pr1ig000 Access 2013 07 03 19 13 34 131999965 0200 Modify 2013 07 03 19 13 34 131999965 0200 Change 2013 07 03 19 13 34 131999965 0200 cat testfile txt file contents Initial conditions Test result The test result matches the expected result Uid and Gid of the file match the user who put the file PRACE 2IP RI 283493 80 23 08 2013 D
136. rods irodsA oldauth iinit Enter your current PAM system password diff q oldauth irods irodsA Files oldauth and irods irodsA differ ils tempZone home niifitest iquest sql ShowPPtimes niifitest tempZone 1209600 01372144219 01372144219 Test result Results match expected results except that ils fails immediately instead of prompting for an iRODS PRACE 2IP RI 283493 64 23 08 2013 D10 2 Second Annual Report of WP10 Detailed tests A valid PAM derived password does exist 3 Initial conditions see PAM derived password does exist 1 username that is niifitest Parameters and input data e zone name that is tempZone cp p irods 1rodsA iexit full Is a 1rods ils cp p irodsA 1rods ils Test procedure Expected result If the user issues an iexit full command then the PAM derived password is removed from irods directory on client side It is expected that ils should prompt for IRODS password and then fail because this user does not have an iRODS password It should work though after a backup of 1rodsA is copied back cp p irods irodsA iexit full Is a irods JrodsEnv ils Enter your current IRODS password rcAuthResponse failed with error 826000 CAT INVALID AUTHENTICATION cp p irodsA irods Is a irods JrodsA irodsEnv ils tempZone home niifitest Test result Results match expected results Detailed tes
137. rt Introatuction 1o CUDA Proqramming amp BSc PRACE News PRACE Summer Shoal Dstave Cech Repubhc PRALCE Tramini events PRACE 5eazannai Scions Weiden tific Visucaltz mi CINEC 18 June 2013 PATE Courses xn Scientific Visualization it A PRACE Whiesepapars PRACE Liver Panniers Trainings Toe s Docume niatan dm Reseanciers IF June 2073 PRACE Summer Goh 2013 Frameworks Tor Solentific Computing intemmiiona Wigkcore i the PRACE Conterences and i on SLOT bers Burmumeer booth mi fSc 13 meetings Hybrid MPUDpenMP programming 12 IDRIS Schon on Exhibaion PRACE industrial las 14 June 2013 Seminars Challenges in DECL11 CAN cosing PRACE Dutreach Events Computation deadine exper 1317141 June 17001 HPC Rieinied vent i Scenes T June 2013 The PRACE Sentit Pree Wise ie 24 zB 26 ar ZB Does wii posa re SS and Sa gies in videa Intemational Summer Scho on HPC Challenges in Cormpadiatiornal Eurapean HPC Net d 8 Jume 2013 Publicis n Crmy Advanced Toots More PRACE News workshop EPCC Presenimnbanz a evenm e FFLACE related news A coordinated Shared Memory Mezsnpe Passing Programing wiih MPi approach fe zcence in Programming wiih T EPCC Europe OpenMP 1 EPCU Figure 6 Event Integration Screenshot 1 PRACE Training Events PRACE 2IP RI 283493 33 23 08 2013 D10 2 Second Annual Report of WP10 PARTNERSHIP FOR ADVANCED COMPUTING Ing 5 Sle PRACE in a
138. s difficult due to the unappropriate documentation Finally the two node cluster exported the file systemking to the client where it could be mounted successfully Although stated by the documentation as available the replication feature did not work Further investigation is required to determine if this is only due to a wrong configuration 3 4 7 Detailed Description of the Performance Measurement Two major questions related to file system performance are of special interest First what 1s overhead introduced by a distributed file system compared to a local one Second how does the distributed file system scale and how does this influence the performance This finally will allow for the comparision of the different distributed file systems PRACE 2IP RI 283493 24 23 08 2013 D10 2 Second Annual Report of WP10 Specific FIO 48 jobs have been used to get answers to these questions where each of the file systems had to undergo all the tests described in the matrix below block size VO type operation type Psie __ 16KB 32KB mad X X X synchronized wie X X X Lmadew e X X X wie X X X Lmadew e X X X Table 4 Matrix of test cases for file systems A simple 4KB synchronized read FIO job for the raw block device looked like this random read sync 4K rw randread Size 512m blocksize 4k directory mnt ext4 In addition four more complicated workloa
139. session information such as mouse keyboard USB and audio across a standard IP network It provides full frame rate 3D graphics and high resolution media The PCoIP protocol encrypts and compresses the data stream on the server side using either dedicated hardware or in software using VMware The data stream is received and decoded PRACE 2IP RI 283493 28 23 08 2013 D10 2 Second Annual Report of WP10 at the receiving end using a stateless zero client or in software VMware View The software solution does not currently support Linux as host operating system The latest generation stateless device supports up to two channels at 2560x1600 or four channels at 1920x1200 and includes VGA DVI and DisplayPort display interfaces The hardware based solution is 100 operating system and application independent The video signal from the graphics card is routed directly to the PCoIP host adapter where it s processed using hardware and transferred to the network using the onboard dedicated GigE NIC Power USB and audio are handled over the PCle bus The SNIC hardware based PCoIP solution consists of two dedicated graphic nodes that is part of the production HPC cluster Alarik The graphic nodes have 32 GB RAM 16 cores 2 sockets and Nvidia Quadro 5000 graphic cards Each node 1s equipped with an EVGA PCoIP host adapter card that ingests the pixel stream s from one or both DVI D outputs of the Quadro 5000 card On the client side curre
140. sfer technologies PRACE 2IP RI 283493 39 23 08 2013 D10 2 6 2 2 DECI peer review tool functionalities requirements list Second Annual Report of WP10 H Functionality Rate Implementation status July 2013 Electronic submission of project proposals Developers ability to programmatically redesign the forms contents and their integration with the internal database Essential Essential Fulfilled Yes Web based ability form design tool to design and change the project submission and evaluation forms Desiderata Mid 2014 Provide users with complete online control of their data application form user data etc and enable them to effectively view and browse their data 1 e applicants can see all their applications response letters and applications status form the portal Essential Fulfilled Assign different roles coordinator of the process evaluator etc and give access to different functionalities 1 e evaluation assignment evaluation process views and data statistical project submission form and evaluation form according to the different privilege level i e evaluators can gain limited access to relevant proposals and TE This would cause different log in views for Applicants Technical amp Scientific evaluators and DAAC staff Essential Ad hoc implementation of user profiles Missing user interfaces Store applicants data project data TE re
141. simulations across disciplines exploiting existing and evolving European e Infrastructure The diagram below presents the steps of a typical simulation workflow and the resources potentially involved and belonging to different e Infrastructures The diagram was elaborated during the pilot activity Job control amp 1 CPMD quantum mechanical simulation parameters for calculating potentials for clay and g GridSpace amp QCG 7 Data movement polymer molecules executed on a local cluster 2 LAMMPS AA atomistic MD simulation to determine accurate microscopic system properties targeting PRACE Tierl resources 3 LAMMPS CG iterative coarse grained simulations intend to find the right potential for the coarse grained system executed on EGI resources 3 LAMMPS CG 4 LAMMPS CG large scale coarse grained EUDAT a simulation to model the final system and oa 4 LAMMPS CG obtain important macroscopic properties PRACE Tier 0 a targeting PRACE Tier0 resources Figure 1 Architectural diagram of the VPH use case pilot A collaboration with this project was already in place but after the EEP EUDAT EGI PRACE workshop it was reorganized and merged under this broader collaboration umbrella The only site currently involved for PRACE is LRZ which hosts the software components QCG Computing MUSCLE QCG Broker QCG Notification necessary to execute MAPPER workflows PRACE 2IP RI 283493 11 23 08 2013 D10 2 Second A
142. straightforward and it could be mounted on the client easily The replication worked seamlessly when both nodes have been online but the write failover from one node to another when a server node went offline during a write stalled the client for more than 10 seconds The write operation was in a deadlock state in the operating system until a timeout occured but then it continued successfully The automatic repair was also working successfully after the offline node joined back online Thus GlusterFS is pretty simple to install and operate all the required features are available but it lacks some agility and tuning options Gfarm Gfarm has an outdated and incomplete documentation and the Debian packages provided in the repositories are built from non current versions Based on such old software a two node file server environment has been successfully setup but the client could not connect to them and thus not mount the filesystem This has been identified as bug 47 which 1s already fixed in a newer version Therfore it was attempted to compile a current version from the source There have been several header and library errors which could not be fixed So it was decided to postpone Gfarm evaluation and no result can be presented yet Coda Coda has an extensive documentation but some of it 1s referring to older versions Since the developers supply Debian packages for Coda software installation went smoothly The configuration wa
143. t Initial conditions CLEANOUT parameter set Test result The files which have been staged in remain in the bin cmd directory and the files defined as INPARAM remain in the stagearea The test result doesn t match the expected result Initial conditions NOVERSION parameter set Test result The NOVERSION parameter stops the versioning of the execution directories During our test the versioning still goes on The test result doesn t match the expected result Initial conditions CHECKFORCHANGE parameter set Parameters and input data the file to check which can be in the WSO environment or anywhere in 1RODS Test result Conclusions Files are not stage in out and the new execution directory is not created Nevertheless the rules and micro services are executed The test result doesn t match the expected result The workflow objects feature provides some interesting mechanisms to help the users to run 1RODS workflows in an integrated environment PRACE 2IP RI 283493 58 23 08 2013 D10 2 Second Annual Report of WP10 The parameter file allows running workflows in a different context and provides a way for the user to easily interact with the IRODS environment We found in this first release several dysfunctions as well as a security issue These issues are going to be reported to the IRODS developers However this feature remains difficult to use and to integrate in the user development envir
144. t Dataset Type Workload Parallel Streams Throughput Mbps 1 4 A 100 files 100 files 5 tro es 8068 ae D e o L C e O UJ O O O NO N 7 JA A 100 files 1000GB 184 75 Bo i I files 1000GB 8 19225 9 A 100files 1000GB 193 10 le el 1000GB 166 27 Qualitative Laan eee Reliability No crashes reported during the tests Footprint Minimal It doesn t require administrative rights Can be installed Intrusiveness by a normal user Maintenance 5 No maintenance required by system administrators a olerance a Hu Good first version released in 2011 last version v2 5 on April Maturity 2013 Acceptance growing Table 10 Example of a filled file transfer evaluation sheet here for bbcp between CINES and CEA 6 5 Performance Measurement of Remote Visualization VirtualGL and TurboVNC setup The tests for RCM VirtualGL TurboVNC have been run on the CINECA infrastructure by opening a session from local TurboVNC client connecting to the remote visualization nodes The frame rate has been measured with tcbench a tool included in the VirtualGL bundle that grabs a small portion of the VNC window and counts the number of different frames in a fixed amount of time In the presence of a running OpenGL application which changes continuously the image this tool seems to properly evaluate the real frame rate Two graphics test applications have been used th
145. t question o start with a reduced IRODS infrastructure and then build up a full operational infrastructure providing high avalailabity as well as a well a defined IRODS environment for each scientific project define the entire system architecture specify the data management policy specify the conventions to be used define the core services users and system oriented access modes and interfaces to the infrastructure ex aO OS x since IRODS is a technology user communities will utilize for their data management as seen in projects like EUDAT it is most likely necessary to support it in PRACE too Also the future development should be influenced to address special HPC requirements Thus this activity 1s planned to be continued in T6 3 of PRACE 3IP PRACE 2IP RI 283493 21 23 08 2013 D10 2 Second Annual Report of WP10 3 4 File System Technologies The sub task File System Technologies 1s part of the task T10 2 Evaluating data services In this section the basic framework for the evaluation of distributed file systems 1s described From possible use cases the technical requirements are derived Then a methodology for testing the measurement metrics initial results and some conclusions are presented Four file systems Coda Gfarm Ceph and GlusterFS have been evaluated These differ greatly in features maturity and operational difficulty but most of them seem to fulfill the requirements for being used as file system being shared betw
146. the default settings IRODS appears here as a simple tool for transferring files allowing good performance Performance tests must be continued including additional test cases on the PRACE high performance 10 Gbit s dedicated network to show the full IRODS performance capabilities since the testbed characteristics mostly did not benefit from the dedicated PRACE network But the performance figures already seen prove iRODS as being an alternative to the standard transfer tool GridFTP even offering additional powerful functionalities Direct Access Resources evaluation summary The Direct Access Resource feature provides a way to have direct access to the files in an IRODS resource through the filesystem they reside on However taking advantage of this feature depends on 1RODS users having identical userids on the machines that host the PRACE 2IP RI 283493 19 23 08 2013 D10 2 Second Annual Report of WP10 filesystem and having sufficient file access rights If the access rights are lacking for a given user the files on the system will be owned by the root user and thus not being accessible Futhermore since the iRODS server must run as root for the direct access resources to work the need for this feature should be carefully weighted against possible security concerns iDrop Graphical Client Interface evaluation summary The focus has been on iDrop features and user experience and not on the setting of the testing environment or its int
147. tions The following table fixes some important definitions related to a file transfer activity that will be considered Measure Definition unit Capacity Link Speed Gbps Link with the lowest capacity along a path see Figure 10 Capacity of the end to end path Capacity of the Narrow Link Utilized Bandwidth Current Traffic Load Available Bandwidth Capacity Utilized Bandwidth Tight Link Tight Link Link with the least available bandwidth in a path see Figure 10 Bandwidth Delay Product BDP The number of bytes in flight to fill the entire path BDP Capacity RTT Table 7 File Transfer Measures Definitions Figure 1 provides an example for determining narrow and tight links of a network path E Narrow Link Nas Figure 10 Graphical example for narrow and tight network links Following the formula stated in Table 7 the BDP for a network with 1 Gbps of capacity and S0ms of RTT is BDP 1000Mbps 0 05s 50Mb 6 25MBytes 6 4 3 Hardware and Configuration Requirements It is assumed that different persons will be involved in the evaluation of different file transfer tools by using different network paths with unpredictable network conditions Defining hardware requirements for the tests 1s a solution in order to mitigate as much as possible the effect of different conditions These requirements have been identified and described in the following sections TCP Buffer Size
148. tomatically the execution within the 1RODS environment preventing the end user to ingest or retrieve from into 1RODS files needed for the execution or produced by the workflow The evaluation consists in building workflows using different set of parameters for checking both the workflow environment creation and execution The workflow object feature is intended to be used by end users However in order to run this feature end users need to be defined as rodsadmin to allow collections to be mounted Responsibility SITE ROLE TASK IDRIS Full evaluation References 1 https www irods org index php Release Notes 3 2 2 https www irods org 1index php Introduction to Workflow as Objects 3 https www irods org index php Workflow Objects 28WSO 29 4 https www irods org index php Realized Objects 5 https www irods org index php glossary 6 https www irods org index php Downloads 7 https bscw zam kfa q uelich de bsew bsew cgi 919814 8 https oscw zam kfa juelich de bscw bscw cgi1 8 19591 Tested Components The workflow object evaluations focus on the software The related documentation is specified at 1 2 3 4 The iRODS release used for the evaluation is 3 2 It can be downloaded at 6 The evaluation requires applying a patch related to the file mssoStructFileDriver c unnumbered for now This patch has been delivered by the IRODS developers and it has to be installed and iRODS has t
149. torage volumes Each of these volumes consists of 100 GB on SATA disks configured as RAID6 The VMs have been configured into two storage nodes and one client 3 4 5 Deployment and Feature Validation Testing Methodology Identical deployment and feature tests have been applied for each file system to achieve as much comparability as possible The following common steps apply to all scenarios e Detailed inspection of the installation documentation e Installation of the latest Debian package available for the given file system from Debian repositories or from the developers themselves e Setup of the software for a two node cluster and a single node client according to the instructions in the manual e Mounting of the file system on the client e Troubleshooting any possible problems and making notes of any difficulties e Proceeding with the evaluation if everything worked so far and the client can read from and write to the file system o Testing replication Writing data once and checking for readability on both nodes Writing data when only one node is online and observing the repair process when the other node comes back online o Testing fault tolerance Testing what is happening if one node from the cluster is becoming unavailable while writing data to it The results for each step have been documented which is summarized in the next section 3 4 6 Test Results Finally four file systems could be tested according to the aforementioned m
150. tre of Excellence for EET HECTOR offers instruction and Newsletters a ee WATCH LIVE HOW Shared Memory Programming with OpenMP EPCC PRACE SUMMER SCHOOL 2013 7 21 uly fava Oze Monday 1 Juy 2013 Shared Memory Programming with OpenMP Almost all modem computers now have a shared memory architecture with multiple CPUs connected to the same physical memory for example multicore laptops or large multi processor compute servers This course covers OpenMP the industry standard for AUTUMN SCHOOL 2013 23 27 September ljubljana Slovenia Message Passing Programming with MPI i EPCC Wednesday 3 Jury 2013 Course Descnption The workf s largest supercomputers are used almost exclusively to run applications which are parallelised using Message Passing The course covers all the basic knowledge required 10 write parallel programs using this programming model and im directly appbcable to almost Syndicated sites Past ATC Cooumez PRACE Seasons Shook Pei PRADE temo TAA etaflop Supercomputer LRZ Eid frg PRACE Tang Dens Parimers Trainings hop gives an introduction to the usage of the new Petaflop class EET ES Supercomputer at LRZ Supa MUC The first three days of this are dedicated to za presentations by Intel on their software development stack compilers tools and E m libranes the remaining day will be comprised of Upcneing PATC courses Upmxcing PRADE Seana Ss Advanced OpenMP EPCC Upceing PRA
151. tructure The objective of task 10 1 1s to identify and evaluate options for technical enhancements to the existing Tier services Input did come from within the work package other work packages like WP2 for the DECI Portal see 2 4 for the handling of the DECI proposals or as a result of surveys as for the Storage Accounting and Reporting see 2 1 2 Furthermore through the collaborations direct user or user community requests as the pilots together with EGI and EUDAT see 2 7 5 influenced the working directions of this task 2 1 Accounting Current accounting covers CPU usage only In this area improvements for the storing of the accounting information and the easy access to 1t for users are a major task In addition with the increasing amount of data produced more and more considerations come up to also include accounting of storage usage 2 1 1 Central Accounting Repository A centralized accounting service was set up in previous years using the Grid SAFE tools developed by EPCC 4 In October 2012 a document was produced as input for the acceptance procedure as a production service Based on this input the members of the operation groups of all sites both Tier 0 and Tier 1 have accepted in November 2012 to propose to PRACE management the Grid SAFE facility as a production service with classification additional as defined by the PRACE Service Catalogue The service 1s included PRACE 2IP RI 283493 2 23 08 2013 D10 2 Second A
152. ts A valid PAM derived password does exist 4 see PAM derived password does exist 1 Initial conditions ShowPPtimes will be used username that is niifitest Parameters and input data zone name that is tempZone PAM asks for the password N n fitest A admin Test procedure e A PRACE 2IP RI 283493 65 23 08 2013 D10 2 Second Annual Report of WP10 iquest sql ShowPPtimes niifitest tempZone N cp p irods irodsA iinit diff q irodsA irods irodsA A iquest sql ShowPPtimes niifitest tempZone Expected result Issuing an iinit command extends the lifetime of a PAM derived password On client side a new irodsA should be created which differs from the old one Modification timestamp should be updated in 1CAT iquest sql ShowPPtimes niifitest tempZone Test result 1209600 01372144219 01372144219 cp p irods irodsA iinit Enter your current PAM system password diff q irodsA irods irodsA Files irodsA and irods irodsA differ iquest sql ShowPPtimes niifitest tempZone 1209600 01372144219 01372144346 The test result matches the expected result Detailed tests An existing PAM derived password is invalidated on client side 1 Initial conditions see PAM derived password does exist 1 Parameters and input data rodsA at client side touch m t 20131231 irods irodsA ils Expected result The ils icommands should
153. ture such as availability accessibility and performance of PRACE resources and services The main goal of this task 1s to address user requirements by providing desired PRACE 2IP RI 283493 7 23 08 2013 D10 2 Second Annual Report of WP10 functionality based on existing or novel technologies The portal team comprises members of three PRACE sites BSC Spain LRZ Germany and PSNC Poland The team is led by LRZ The task started with an evaluation of existing and operational technologies that could be used for providing PRACE end users with required information Throughout the evaluation process no suitable existing service could be identified which led to the conclusion that the information portal has to be developed based on a novel technology The architecture of the information portal is largely based on standards to ensure easy integration with PRACE operational services Some of the standards adopted or implemented in the portal include OGF GLUE2 0 24 REST 25 and JSR268 26 For the first release of the portal the following five high priority requirements were chosen for implementation e PRACE Link Availability Internet Link Availability Core Services Status Account Usage System Information At this moment the first implementation phase 1s nearing its completion Implementation of necessary information providers is finished logic and web interface are currently under development PSNC has designed and implemented the f
154. uired development related activities which were successfully performed in WP 10 PRACE 2IP RI 283493 3 23 08 2013 D10 2 Second Annual Report of WP10 2 2 1 Initial Status PRACE RI Web Site The PRACE RI main website is based on SPIP CMS 21 It is hosted at and administered by CINES The software has a GPL license and documentation 1s mostly 1n French It 1s written in PHP with a MySQL database New features can be added as plugins through a website backend available to web administrators or by storing it directly on the machine available only to CINES staff PRACE Events System PRACE Events System is based on InDiCo software 22 InDiCo is a web application for scheduling and organizing events from simple lectures to complex meetings workshops and conferences with various sessions and contributions It was originally developed in the framework of the EU InDiCo project 22 but currently InDiCo is free software licensed under terms of GNU General Public License GPL The InDiCo user guide can be found at 23 The PRACE Events System is hosted at and administered by IPB 2 2 2 Technical Background Integration The aim 1s to enable automatic display on the PRACE RI website of upcoming events that are entered in the InDiCo System Events should be displayed in three categories e Upcoming events sorted by date first to come is on top e Past events sorted by date latest on top e Calendar view all events sorted by year
155. unts on the host machine with the same username and password in order to have access to files through the filesystem SITE TEST ENVIRONMENT Scientific Linux 6 3 virtual machine on PARADOX 2 Proc Intel Xeon CPU E5345 2 33GHz 2 GB RAM 1Gb Ethernet card Testing Methodology The test will focus on the functional aspect So we will check whether the functionality is provided that if whether it does what is supposed to do the easiness of use the reliability and maintainability Tests Description Direct access resource creation irods user account in 1RODS defined as rodsadmin irods service running as root with DIRECT ACCESS VAULT configuration enabled MySQL used for ICAT database started independent of 1RODS or an independent PostgreSQL instance must be used because default Initial conditions PostgreSQL that is bundled with 1RODS can not be started as root alternatively when starting the server as root use istart command instead of start passed to irodsctl script Direct access resource at path opt rodshare a dummy text file used to verify that resource 1s accessible for file operations Parameters and input data PRACE 2IP RI 283493 79 23 08 2013 D10 2 Second Annual Report of WP10 cd 9IRODS HOME sudo irodsctl start iinit iadmin mkresc testresc direct access filesystem cache irods ipb ac rs opt rodshare testresc resource should be created on the irods instance Expected
156. vailable network performance The possibility to enable secure USB bridging to the host system opens up interesting options for transferring data and connecting other interaction devices An administrator can disable this function if needed PCoIP is a commercial solution using proprietary hardware both on server and client side something that somewhat limits the usage for academic purposes even if the price level is very decent especially when put into a performance and image quality context Performance wise the resulting image quality and interactive performance is perceived as very good and predictable when running on the campus network using a 1920x1200 resolution The technology adapts to different network situations in a user controllable fashion to allow either automatic adjustments or using fixed numbers such as maximum peak bandwidth allowed and how the system should behave during congestion Stateless means there is no record of previous interactions and each interaction request has to be handled based entirely on information that comes with it PCoIP encodes each pixel to a lossless state once they stop changing to ensure a pixel perfect image PRACE 2IP RI 283493 29 23 08 2013 D10 2 Second Annual Report of WP10 The bandwidth needs depend on the frame content spatial resolution number of display channels and other communication such as audio and USB The largest contribution to the bandwidth usage is the portal pixel tra
157. view data suggested extra TE info SE data ranking info etc into the DECI Database Create and or change user s evaluator s site s countries info Essential Essential Postponed Postponed Support the process of submitting a short report from the PI after the completion of the project the template of this report being downloadable from the tool Essential Not yet schedulled 10 Create statistics reports of the DECI process 1 e number of technical evaluations per site number of scientific evaluations per evaluator Moreover the publications related to work done with DECI resources should be tracked via the proposed tool Copy or link the relevant data from the web based tool when needed into the DPMDB i e project name home site technical requirements such as CPU type number of jobs memory simulation codes etc Desiderata Essential Not yet schedulled Not yet schedulled 11 Copy summary of projects resource usage from DPMDB to the web based tool so that PIs can view accounting information related to their projects without learning a new tool DART Desiderata Not yet schedulled 12 Create and export documents and information that should feed other systems or processes 1 e automatic generation Desiderata Partial implementation PRACE 2IP RI 283493 40 23 08 2013 D10 2 Second Annual Report of WP10 Functionality
158. with PRACE s central LDAP Notably for external authentication of DECI staff users e Interfacing with the Grid SAFE DART accounting infrastructure for follow up of awarded projects resources consumption Some further DECI specific processes might need some additional supporting functionality in the tool such as the DECI Access and Allocation Committee DAAC where awarded proposals are assigned to sites and actual computer systems 2 5 PRACE Information Portal PRACE users require various information to efficiently use services offered in the e Infrastructure This information includes network status and performance HPC resource maintenance schedule service availability and functionality and so on PRACE operates a number of tools e g Iperf for network and Inca for service monitoring to measure and collect detailed statistics on availability and functionality of production services This data 1s to a large extent available solely to PRACE staff members as raw data sets often contain security or privacy relevant information This limits the sharing and distribution of e Infrastructure state data among end users Annual PRACE user surveys emphasize these limitations and provide details on the kind of data users need to efficiently work in PRACE A sub task of WP10 is responsible for the evaluation design and implementation of a portal called the PRACE Information Portal for providing users with information on the PRACE e Infrastruc
159. y image compression schemes to trade off image quality for frame rate usually on single images as the more effective interframe compression schemes used in video streaming generate excessive latency However this loss in 1mage quality 1s really difficult to measure in a quantitative way as it heavily depends on image content itself In order to quantify latency a proper setup is needed high speed camera and the procedure can be significantly time consuming as described in an article The truth about latency in Cloud Gaming 49 furthermore since latency is mostly dominated by the network components it can be highly variable depending on the client server network load In order to quantify the frame rate a tool tcbench included within the VirtualGL distribution which adopts a simple but effective approach has been used The tool runs on the client machine and inspects a small portion of the VNC window detecting how many times the screen changes per second If an application is run which constantly changes the screen then the tool correctly detects the screen change and computes the real perceived frame rate disregarding frame spoiling techniques Regarding which application is used for testing two approaches are possible the first 1s to use a very simple and fast graphic application to minimize the application overhead to be sure of being limited by just the grab compression transport decompression involved in remote visualization

D10.2

Contents

Download Pdf Manuals

Related Search

Related Contents