Home

IST-FP6-003769 CATNETS D2.3 Annual Report of WP2

Contents

1. Column description time stamp in milliseconds long ID of the agent String name of the metric String value start or end entry Negotiation ID String result of the negotiation failure or success only written for end entries e Filename basic_service_provisioning_time txt Description Time needed for basic service provisioning This includes execution time and successful and failed negotiations Sample entry 1178575207841 CSA6 Site9 basic_service_provisioning_time 3146 success Column description time stamp name of the agent name of the metric time needed for allocation failure success e Filename BS_R_Mapping txt Description Mapping of service market negotiation to resource market negotiations Sample entry 1178575203777 BSA1 Site9 BS_R_ Mapping Negotiation0 Negotiationl Column description time stamp name of the agent name of the metric negotiation id of the service market negotiation list of negotiation ids of the related resource negotiations e Filename catallactic_initial_price_range txt Description Initial price range of agent Sample entry CHAPTER 2 SIMULATOR REFINEMENT 25 1178575383665 CSA0 Site0Buyer catallactic_initial_price_range bs1 104 41305177623482 134 41305177623482 1178575383685 CSA19 Site25Buyer catallactic_initial_price_range bs1 93 78254079037917 123 78254079037917 1178575383755 RAO Site0Seller catallactic_initial_price_range rl 112 4561961146844 142 45619
2. 21 Information Society os Technologies IST FP6 003769 CATNETS D2 3 Annual Report of WP2 Contractual Date of Delivery to the CEC 31 08 2007 Actual Date of Delivery to the CEC 24 09 2007 Author s Werner Streitberger Universitat Bayreuth Torsten Eymann Universitat Bayreuth Floriano Zini RST Fondazione Bruno Kessler Trento Bj rn Schnizler Universitat Karlsruhe Hong Tuan Kiet Vo Universitat Karlsruhe Workpackage WP2 Simulation Est person months 16 Security public Nature final Version 1 0 Total number of pages 49 Abstract This document describes the activities performed in WP2 Simulation in the third and final year of the CATNETS project In particular it focuses on WP2 s main task namely Simulation of application layer networks and refinement Keywords optional CATNETS simulator user guide simulation environment refinement CATNETS Consortium This document is part of a research project partially funded by the IST Programme of the Commission of the Eu ropean Communities as project number IST FP6 003769 The partners in this project are LS Wirtschaftsinformatik BWL VID University of Bayreuth coordinator Germany Arquitectura de Computadors Universitat Politecnica de Catalunya Spain Information Management and Systems University of Karlsruhe TH Germany Dipartimento di Economia Universita delle Marche Ancona Italy School of Computer S
3. File Formats V BRITE OTTER Use Java Exe w Build Topology Figure 2 3 Automated scenario generator Catnets2 parameter tab Links inv The site for the agent is chosen with probability inverse proportional to the number of site links The more the site is connected the smaller is the probability to host agents Dist dir The site for the agent is chosen with probability proportional to the dis tance between the site and a pivot site the more the distance the greater the probability Dist inv The site for the agent is chosen with probability inverse proportional to the distance between the site and a pivot site the less the distance the greater the probability Percentage of BSAs providing a specific types of basic service The user can decide the percentage of agents providing a specific type of basic service By clicking on the Create Tables button the predefined percentages are shown and the user can modify them Percentage of RAs providing a specific available resource bundles The user can de cide the percentage of agents providing a specific ARB By clicking on the Create Tables button the predefined percentages are shown and the user can modify them CHAPTER 2 SIMULATOR REFINEMENT 10 2 1 2 How to generate a scenario Once all the parameters have been set a folder has to be selected for the output files A click on the button Build Topology generates the scenario Four input files fo
4. negotiation Id seller basic service e Filename util_satisfaction_buyer_resource_central txt Description Utility metric of a basic service on the resource market Sample entry 20983 Negotiation80 BSA0D Site9 util_satisfaction_buyer_resource_central 0 remove 21053 Negotiation108 BSA1 Site3 util_satisfaction_buyer_resource_central 0 9222974262273637 success Column description time stamp negotiation Id name of the writing agent name of the metric satisfaction value remove success remove if the order was withdrawn due to a time out success otherwise Note Only used in the central case e Filename util_satisfaction_buyer_service_central txt Description Utility metric of a complex service on the service market Sample entry 23747 Negotiation117 CSA5 Site20 util_satisfaction_buyer_service_central 0 8951489201843724 success 26071 Negotiation109 CSA8 Site24 util_satisfaction_buyer_service_central 0 remove Column description time stamp negotiation Id name of the writing agent name of the metric satisfaction value remove success remove if the order was withdrawn due to a time out success otherwise Note Only used in the central case CHAPTER 2 SIMULATOR REFINEMENT 32 e Filename util satisfaction resource buyer decentral txt Description Resource buyer s utility Sample entry 1178488804195 BSA0 Site23 util_satisfaction_resource_buyer_decentral Negotiationl 0 21748918596452
5. Chapter 3 Guide to conducting simulations This chapter describes first how to install and run the CATNETS scenario generators and simulator and then how to set the various parameters of the simulation 3 1 Installation and Running These instructions describe how to install and run CATNETS scenario generators and the CATNETS simulator Three ZIP files of the needed packages can be downloaded from the CATNETS web site http www catnets org download The first ZIP file called catnets manual scenario generator zip includes the source code the the generated class files and the javadoc doc umentation of the manual scenario generator The second ZIP file called catnets automated scenario generator zip includes the source code the generated class files and the javadoc documentation of the automated scenario genera tor The third ZIP file called catnets simulator zip includes the source code the generated class files the javadoc documentation and usage examples of the simulator 3 1 1 Software Dependencies e Java TM 2 Platform Standard Edition Development Kit 5 0 or greater e Ant Version 1 6 5 or greater to re build the code after any changes or to run func tional tests 34 CHAPTER 3 GUIDE TO CONDUCTING SIMULATIONS 35 3 1 2 CATNETS manual scenario generator Installing 1 Start by unpacking the code from the ZIP file by using an archiving package such as WinZip do not open the file directly from the browser bu
6. after they ve sent cfps timeout Time in milliseconds agents wait for non blocking reception of messages dur ing negotiations for services or resources message size Size of messages in Kbytes size 0 implies instantaneous message delivery CHAPTER 3 GUIDE TO CONDUCTING SIMULATIONS 42 3 2 5 Other parameters time advance Use advanced grid time yes or not no see Section 2 2 for details time of day The time of day used as starting point Should be in hours with minutes after the decimal point e g 22 5 for 22 30 and must be on the hour or half hour metrics path The path where files recording metrics collected during simulations are stored 3 3 Simulation output Every simulator run produces the set of technical metrics described in Section2 5 One file is generated for every metric the file name being lt metric_name gt txt All the files produced by a run are stores in a newly created directory whose name is the result of the Java call System currentTimeMillis which returns the current absolute time in milliseconds Chapter 4 Conclusions 4 1 Achieved results The CATNETS simulation environment poses low requirements on simulation run Im plemented in pure JAVA the CATNETS simulator runs on all machines which are sup ported by the Java Runtime Environment This enables small scale simulation on desktop machines and large scale simulation on multi core 32 and 64 bit server machines The pur
7. by the manual or automated scenario generator These files are expected to be in the root directory of the market property generator The process of generating the market property file is divided into two steps which are illustrated by two screens respectively Within step one tradable products on the resource market are specified In a second step pricing limits are set for each basic service type and a resource product is assigned to each basic service type An outline is given for both steps which are illustrated by an example in the following Specification of Products Starting the MPG will result in the screen depicted in Figure 2 4 This screen is divided into four different regions Within region 1 all possible product types are shown in a drop down box By selecting one of these types the user is enabled to instantiate a product type by pressing the add button on the right Every instantiated product will appear in region 2 Instances of the same type are distinguishable by means of a number behind the identifier Marking a single instance of a product type in region 2 makes it possible to customize it within region 3 Here all product specific options are addressable After customizing the products listed in region 2 the next button in section four hast to be pressed Figure 2 4 shows a snapshot of MPG which uses the example data provided below Two different product types are instantiated storage one instance and cpuram two in stan
8. latency 11 2 25 Event driven time model y 2 2 amp avai A eS are AA ee 11 2 3 Centralised Market e ASA SO ghee ASE Pak d 12 24 R ONAN ACHE market A AAA AE RA ee 13 2 4 1 Configuration of the Catallactic market 13 2 4 2 Market Property Generator 16 2 4 3 Service and Resource Discovery 18 2 4 4 Configuration of the catallactic strategy and learning algorithm 21 2 4 5 Simplified Catallactic strategy 23 2 9 Metrics od ir A A A A Ende S 23 3 Guide to conducting simulations 34 3 1 Installation and Running 00000202 34 3 1 1 Software Dependencia Sa Wk GOR Be A ER 34 3 1 2 CATNETS manual scenario generator 35 3 1 3 CATNETS automated scenario generator 35 3 1 4 CATNETS simulator AR ROR ROR ee 36 3 2 CATNETS simulator parameter MS ociosa o 0 te aa 37 3 2 1 General Parameters 4 x Sw a aa tw te we oe St ee Sk 37 3 2 2 Central Market Parameters a Re eR 38 3 2 3 Market Parameters o oo aaa e 40 3 2 4 Negotiation Parameters a 41 3 2 5 Other parameters os a e a 42 3 3 Simulation output co der o e 42 CONTENTS 2 4 Conclusions 43 4 1 Achieved results 2 6 6 ec ee ee a ee ee 43 4 2 Current Limitations 0 0 0 0 00000000 2 eee 44 4 3 Future Extensions 45 Chapter 1 Introduction The objective of WP2 Simulation of the CATNETS project was to ide
9. to his basic service type Figure 2 5 Screen two of MPG 2 4 3 Service and Resource Discovery Large scale Grids will borrow some of the characteristics of today s P2P systems in re source participation unreliable resources and intermittent participation will constitute a significant share This environment is likely to scale to thousands of resources shared by thousands of participants no central global authority will have the means and the in CHAPTER 2 SIMULATOR REFINEMENT 19 centive to administrate such a large collection of distributed resources The nodes offer resource bundles as a service or application services basic services and complex ser vices that other nodes services want to use An issue central to such Grids is how to locate a service in the distributed system that provides it There are various approaches to service discovery In CATNETS we consider a decentralized discovery mechanism based on flooding With flooding a node that wants a particular resource or service con tacts all neighbours in the system which in turn contact their own neighbors until a node that provides the requested service is reached Flooding enables resource discovery with out directories or knowledge about the specific topology of the system thus offering an attractive mechanism for resource discovery in dynamically evolving systems In abstract t
10. too complex to analyze and to identify specific effects of the implemented auction mechanisms This originates in the fact that too many parameters of the model influenced the overall outcome As a consequence the reasons for particular outcomes cannot be easily identified In order to observe and identify specific effects of the auction mechanisms a simpler bidding model was developed in the third project year Valuation and reservation prices of agents are drawn from a normal distribution where mean and standard deviation can be set by means of a configuration file As the simulation scenarios show the simpler model allows a more profound analysis of the metrics to be observed Both models can be selected by means of a configuration file see Section 3 2 2 for details CHAPTER 2 SIMULATOR REFINEMENT 13 2 4 Catallactic market The components of the Catallactic strategy presented in Deliverable D1 1 and formalized in Deliverable D2 1 were optimized during the simulation of several test scenarios in project year 3 A configuration file was introduced for defining the traded products on the Catallactic market which enables the configuration of the catallactic market properties see subsection 2 4 1 Besides the configuration of the Catallactic market properties all parameters of the Catallactic strategy could be configured with configuration files too Subsection 2 4 2 presents a visual tool for generating the market properties which ease
11. types a P2P messaging layer with flooding load link dependent message delay and a simple message failure model This enables the simulation of real life influences on the resource allocation approaches Interfaces ease the implementation of an improved P2P layer or new message types Various tools were developed to support the configuration of the simulator Scenario generators help to create new scenarios or configure the catallactic market The simu lator supports plain text file based configuration This allows fast reconfiguration of the scenarios between simulations runs A large set of metrics was implemented in the simulator This set of metrics helps to debug and evaluate the simulation runs Technical and economic metrics are written into text files which are evaluated with MATLAB scripts ex post Again the use of text files gives the possibility to use any tool for analysis The simulation environment and all developed tools will be released under an open source licence This will give other researchers the possibility to modify and extend the simulator for their own research 4 2 Current Limitations The simulator and its current implementation of the CATNETS scenario have some lim itations These limitations result from assumptions made to reduce the implementation complexity Currently the allocation approaches don t support parallel negotiations Seller and buyers can negotiate with only one other negotiation partner at t
12. using a uniform random distri bution 3 RandomWalkUnitaryAccessGenerator CSs are accessed using a unitary random walk starting from a CS chosen using a uniform random distribution 4 RandomWalkGaussianAccessGenerator CSs are accessed using a Gaus sian random walk starting from a CS chosen using a uniform random distri bution random seed Determines whether the seed used by various methods within the CAT NETS simulator where random numbers are required is fixed or random If this is set to yes it will be random if no it will be fixed For example if it is yes a different set of CSs will run each time the simulation is run If it is no the same CSs will run each time max queue size The maximum number of CSs the CSA will hold in its queue before it refuses to accept any more bs execution time The time in milliseconds for a CSA to execute each BS 3 2 2 Central Market Parameters The parameters for the auctioneers are stored in the properties file market_central properties In the following the semantics of each pa rameter is briefly discussed Basic Service Agent Parameters These parameters affect the behavior of a basic service agent basic useServiceMarketPrice Defines the pricing model to be used e basic useServiceMarketPrice 1 Use the price on the service market as a maximum bid for the resource market e g bought ServiceA for 10 bid at most 10 on the resource market e basic useServiceMarketPri
13. 227 1178488805317 BSAOGSite28 util_satisfaction_resource_buyer_decentral Negotiation3 0 0013210052001667583 Column description time stamp name of the writing agent name of the metric negotiation id utility Note Only used in the catallactic approach e Filename util_satisfaction_resource_seller_decentral txt Description Resource seller s utility Sample entry 1178488808659 RA0D Site28 util_satisfaction_resource_seller_decentral Negotiation9 0 10714285524219869 1178488810254 RAO Sitell util_satisfaction_resource_seller_decentral Negotiationll 0 0795453827887711 Column description time stamp name of the writing agent name of the metric negotiation id utility Note Only used in the catallactic approach e Filename util_satisfaction_seller_resource_central txt Description Utility metric of a resource service on the resource market Sample entry 20983 Negotiation85 RAO Site3 util_satisfaction_ seller resource central O remove 21053 Negotiation112 RAO Site3 util_satisfaction_seller resource central 0 08424893267942524 success Column description time stamp negotiation Id name of the writing agent name of the metric satisfaction value remove success remove if the order was withdrawn due to a time out success otherwise Note Only used in the central case CHAPTER 2 SIMULATOR REFINEMENT 33 e Filename util_satisfaction_seller_service_central txt Description
14. 611468442 Column description time stamp name of the agent and role Buyer and seller name of the metric type of the requested service initial lower limit initial upper limit Note Only used in the catallactic approach e Filename catallactic strategy BSA buyer txt catallactic strategy BSA seller txt catallactic_strategy_CSA txt catallactic_strategy_RA txt Description Genotype prices and internal status values of the strategy Sample en try 1178575229131 CSA9 Sitel2 catallactic_strategy_CSA Negotiation59 0 10000000149011612 0 6000000238418579 0 30000001192092896 0 9990000128746033 0 5 0 0 0 bs1 101 09558068467088 111 14513166071382 96 14513166071382 126 14513166071382 Column description time stamp name of the agent name of the metric negotia tion id acquisitiveness price step price next satisfaction weight memory average profit learning generation product id price estimated market price lower limit upper limit Note Only used in the catallactic approach e Filename cfps txt Description Number of received call for proposal messages for each seller agent Sample entry 1178575208583 RAO Sitel5 Seller cfps 5123 Column description time stamp name of the agent role name of the metric num ber of call for proposal messages received Note Only used in the catallactic approach e Filename cfps sent txt Description Sent call fo
15. TE user manual MLMBOla BRITE was extended in order to obtain an automated scenario generator which is able to generate scenarios compliant to the ALN model presented in Deliverable D2 2 WP206 2 1 1 Extensions Two tabs have been added to the GUI of BRITE They are called Catnets1 and Catnets2 and are shown in Figure 2 2 and Figure 2 3 respectively Catnets1 Tab Tab Catnets1 enables the configuration of complex service types basic service types and resource types which equal the products available in the application layer network The following parameters are included in the tab CHAPTER 2 SIMULATOR REFINEMENT 7 Boston University Representative Internet Topology Generator BRITE AS Topology Paramete HS Ls Model waman Model Specific Parameters Node Placement iw Growth Type Incremental x iw Pref Conn Bandwidth Distr Uniform Export Topology Location File Format s V BRITE J OTTER Use Java Exe w Build Topology Figure 2 1 BRITE topology selection Resource parameters The user can define Res the number of resource types avail able in the scenario and Qmaz the maximum quantity for each resource type Res resources will be created each of which having a quantity defined by a ran dom integer in the range 1 Qmaz ARB parameters The user can define MaxRes the maximum number of resources per Available Resource Bundle and 4ARB the number
16. Utility metric of a basic service on the resource market Sample entry 26071 Negotiation111 BSAO Site3 util_satisfaction_seller_service_central O remove 26842 Negotiation135 CSA1 Site8 util_satisfaction_seller_service_central 0 8700211740132909 success Column description time stamp negotiation Id name of the writing agent name of the metric satisfaction value remove success remove if the order was withdrawn due to a time out success otherwise Note Only used in the central case e Filename util_satisfaction service buyer decentral txt Description Service buyer s utility Sample entry 1178488804234 CSA14 Sitel7 util_satisfaction_service _buyer_decentral Negotiation3 0 19718347083943188 1178488807691 CSA6 Site9 util satisfaction _service buyer _decentral Negotiation7 0 07608043102226603 Column description time stamp name of the writing agent name of the metric negotiation id utility Note Only used in the catallactic approach e Filename util_satisfaction_service_seller_decentral txt Description Service seller s utility Sample entry 1178488803219 BSA0GSite23 util_satisfaction_service_seller_decentral Negotiationl 0 0 1178488804211 BSA0 Site28 util_satisfaction_service_seller_decentral Negotiation3 0 1668412999533775 Column description time stamp name of the writing agent name of the metric negotiation id utility Note Only used in the catallactic approach
17. ances to network nodes The following parameters can be defined Allocation mechanism The user can set the allocation mechanisms catallactic or cen tralised for which the scenario is being created CS schedule The user can decide if all the Complex Service Agents can run all complex services or if CSAs can run only a subset which is randomly chosen Agents definition and distribution The user can define Agents which is the total number of agents to be placed in the ALN the percentages of complex service agents C S As basic service agents B SAs and resource agents RAs and the distributions of probability used for the placement of agents in the ALN The following distributions are available Uniform The site for the agent is chosen using uniform probability distribution Links dir The site for the agent is chosen with probability proportional to the number of site links the more the site is connected the greater the probability it host agents CHAPTER 2 SIMULATOR REFINEMENT 9 Boston University Representative Internet Topology Generator BRITE E Topology Type i Level AS ONLY Allocation Mechanism Gatallactic lv Distribution of Probability agents COSAS CBAs 1 unitorm w CSschedule BSAS BSAS t uniform v Jal i RAS RAS 1 uniform iy es Tame LOreate tables CARB Table _ARBIDs Percent aj gt Export Topology Location o Browse
18. bundles and price ranges and shares the intended meaning with the basic service In the service market configuration the basic service consumer defines which of these resource bundles can be bought to fulfill his resource demand on the resource market A resource provider defines its resource product out of a set of basic resources and chooses an arbitrary name for the resource bundle The resource bundle determines the base units of each resource A resource consumer requests a multiple of these CHAPTER 2 SIMULATOR REFINEMENT 15 resource bundle base units Like in the service market configuration the same ini tial resource market price s has to be configured along with the initial minimum and maximum prices for buyers and seller which trade this product and hard upper and lower limit prices The lower limit price is the minimum price a resource provider is willing to sell this product and the hard upper limit price is the budget of the re source consumer Finally each resource provider reads the list of possible resource products arb itemids and decides according to his available resource items if he is able to provide this resource product lt resource_product_id gt resourceids lt resource gt lt resourcen gt lt resource_product_id gt baseunit lt resource gt lt int gt lt resource_product_id gt baseunit lt resource gt lt int gt lt resource_product_id gt seller minPrice lt double g
19. ce 0 Draw the valuation for the resource market from an independent distribution Service Market Parameters These parameters affect the behavior of a service market auctioneer service kprice k 0 1 value for the k pricing schema on the service market Set this to 0 5 in most cases CHAPTER 3 GUIDE TO CONDUCTING SIMULATIONS 39 Resource Market Parameters The following parameters affect the behavior of a resource market auctioneer resource kprice k 0 1 value for the k pricing schema on the resource market Set this to 0 5 in most cases resource numberattributes The number of attributes each resource has resource updateunsuccessful Defines if valuation and reservation prices should be up dated e resource updateunsuccessful 0 Valuation Reservation of unsuccessful agents are not changed after each clearing period e resource updateunsuccessful 1 The Valuation reservation price of unsuc cessful agents should be updated after each clearing period resource orderbook finddisjunctivesets A boolean value that defines if the optimiza tion engine should search for disjunctive order sets on the resource market resource orderbook split This is a fixed value which has to be 0 resource allocator model This is a fixed value which has to be 3 resource allocator solver Determines the external solver to be used e resource allocator solver 0 Use CPLEX to solve the winner determination problem on the resource market e r
20. ces The screen shot shows the available option for product storage 0 Available resource bundles arb cpu 30 arb ram 40 storage 80 Basic services bs1 bsl silver storage 10 bs2 bs2 silver cpu 20 ram 88 bs3 bs3 silver cpu 22 ram 22 CHAPTER 2 SIMULATOR REFINEMENT Market Properties Generator Products Seller min price Seller max price cpu baseunit ram baseunit Buyer min price 75 Buyer max price 1 20 Hard lower limit price 40 Hard upper limit price 175 storage baseunit 5 Figure 2 4 Screen one of MPG 17 CHAPTER 2 SIMULATOR REFINEMENT 18 Products and basic services The second screen of MPG is depicted in Figure 2 5 which is divided into two regions In region 1 every basic service type is enumerated with its specific options Each of those options has to be changed by the user The final option is presented in a drop down box This box assigns a resource product to each particular type of basic service If the box is empty no resource product has been defined previously Pressing the previous button allows the user to proceed on screen one The generate button produces the catallactic market properties file Figure 2 5 shows the configuration for the example data set In that specific case there are two different products which could serve bs2 needs The user needs to choose which one he wants to assign
21. cience and the Welsh eScience Centre University of Cardiff United Kingdom Automated Reasoning Systems Division IRST Fondazione Bruno Kessler Trento Italy University of Mannheim Germany University of Bayreuth LS Wirtschaftsinformatik BWL VII 95440 Bayreuth Germany Tel 49 921 55 2807 Fax 49 921 55 2816 Contactperson Torsten Eymann E mail catnets Ouni bayreuth de University of Karlsruhe Institute for Information Management and Systems Englerstr 14 76131 Karlsruhe Germany Tel 49 721 608 8370 Fax 49 721 608 8399 Contactperson Daniel Veit E mail veit iw uka de University of Cardiff School of Computer Science and the Welsh eScience Centre University of Caradiff Wales Cardiff CF24 3AA UK United Kingdom Tel 44 0 2920 875542 Fax 44 0 2920 874598 Contactperson Omer F Rana E mail o f rana cs cardiff ac uk University of Mannheim Chair of E Business and E Government L9 1 2 68131 Mannheim Germany Tel 49 621 181 3321 Fax 49 621 181 3310 E mail veit uni mannheim de Universitat Politecnica de Catalunya Arquitectura de Computadors Jordi Girona 1 3 08034 Barcelona Spain Tel 34 93 4016882 Fax 34 93 4017055 Contactperson Felix Freitag E mail felix ac upc es Universita delle Marche Ancona Dipartimento di Economia Piazzale Martelli 8 60121 Ancona Italy Tel 39 071 220 7088 Fax 39 071 220 7102 Contactperson Mauro Gallegati E mail gall
22. d model may be desirable for demon stration or debugging purposes The event driven time model is implemented by using a thread called TimeAdvancer which controls the advance of time The thread runs as a lowest prior ity thread which only starts working in general when all other threads are sleeping When it runs it continually checks to see if all other threads are sleeping If there are all threads sleeping it finds out the thread which should be activated again It advances time to this point and wakes up the thread 2 3 Centralised market In the second project year we conceptualized a model for generating values for the bids in the central case WP106 The purpose of the model was to determine what and when an agent bids in the auctioneer cases What denotes the valuation and the reservation prices of agents i e the maximal price which an agent is willing to pay for a service resp the minimum price an agent has for selling a service When denotes the timing of bids i e which event induces an agent to bid for a service Both cases are different in the central and decentral scenario As such it was important to find concepts that are applicable for both scenarios and thus make the results comparable As a result a rather complex model was developed with the aim of imitating the co evolutionary learning algorithms WP 106 which is used in the catallactic allocation approach Preliminary simulation runs evinced that the model was
23. e JAVA implementation gives the possibility of easy adoption of the current code to new simulation scenarios of application layer networks in utility computing or autonomic computing areas The high resource and service abstraction and the two tiered market implementation supports various areas of application The introduction of resource bundles enables the modeling of not only a specific resource type like data resources but complex resource products for future peer to peer enabled Grid applications We assume visualization tech niques and local resource managers are in place and offer an abstract resource bundle to services In CATNETS we implemented two different allocation policies for these re source bundle a centralized auctioneer using a multi attributive combinatorial auction and the catallactic approach using bilateral bargaining A resource provider can select between two implemented resource models for his re source service a shared and a dedicated resource model Using the dedicated resource model super computing or autonomic computing can be simulated whereas the shared resource model represents scenarios of the utility computing field Both resource models allow co allocation of resource bundles from different resource providers This allows the simulations of basic services with high resource demands like the execution of batch jobs in the super computing area Not all possible co allocation scenarios are supported by the curren
24. egati dea unian it IRST Fondazione Bruno Kessler Trento Via Sommarive 18 38100 Povo Trento Italy Tel 39 0461 314 314 Fax 39 0461 302 040 Contactperson Floriano Zini E mail zini itc it Changes Version Date Author Changes 0 1 20 05 07 Werner Streitberger Draft structure 0 2 03 07 07 Floriano Zini Written introduction and added draft material to other sections 0 3 04 07 07 Floriano Zini Completed sections on automated scenario generator and infrastructure 0 4 10 07 07 Floriano Zini Added draft material to section on user guide 0 4 1 12 07 07 Floriano Zini Minor refinements 0 5 31 07 07 Bjoern Schnizler Description of the parameters for the auctioneers and Werner Streitberger the catallactic market 0 5 1 31 07 07 Floriano Zini Re alignment with version 0 4 1 0 6 08 08 07 Floriano Zini Updated user guide 0 7 30 08 07 Werner Streitberger Flooding performance added 0 8 06 09 07 Werner Streitberger Chapter 4 added 1 0 24 09 07 Werner Streitberger Final editing Contents 1 Introduction 3 2 Simulator refinement 6 2 1 Automated scenario generator 0000 eee eee 6 ZL SISRIEMSIOMS Bante Arete od ad dro o dr wk rd G55 6 2 1 2 How to generate a scenario laa a A oe a 10 22D Infr stfUctUre 2 5 2 46 97 eS A A 10 2 2 1 Message delivery failure so waded owe Pare 2 aS 10 2 2 2 Link load dependent message
25. el How to select proposals 0 fifo one shot 2 best price one shot The fifo selection model takes the fastest received proposal and starts the negotiation whereas the best price policy ranks all received proposals and selects the cheapest one for negotiation max coallocation Maximum number of co allocated resources This property defines the number of different resource providers a basic service is allowed to allocate market central service clear Clearing policy for the centralised service market 1 Call Market or 2 Continuous market central service clearinterval Call market clearing interval for the centralised service market defines after how many ms the market will be cleared market central resource clear Clearing policy for the centralised Resource Market 1 Call Market or 2 Continuous market central resource clearinterval Call market clearing interval for the centralised resource market defines after how many ms the market will be cleared 3 2 4 Negotiation Parameters The following parameters regulate how negotiations are conducted in both the catallactic and centralised mechanisms cfp_ann hop count Regulates the propagation of cfps or announce messages over the network when the catallactic mechanisms is adopted learning hop count Regulates the propagation of learning messages over the network when the catallactic mechanisms is adopted discovery timeout Time in milliseconds agents wait for proposal
26. ent techniques will be introduced This will enable the simulation of different quality of service levels on the resource and service market Bibliography MLMB01a MLMBO1b WP106 WP206 WP405 WP406 A Medina A Lakhina I Matta and J Byers BRITE Universal Topology Generation from a User s Perspective April 2001 http www cs bu edu brite user_manual BritePaper html A Medina A Lakhina I Matta and John Byers BRITE An Approach to Universal Topology Generation In Proc of the International Workshop on Modeling Analysis and Simulation of Computer and Telecommunications Systems MASCOTS 01 Cincinnati Ohio USA August 2001 WP1 Annual Report of WP1 Technical Report WP1 D2 CATNETS EU IST FP6 003769 2006 WP2 Annual Report of WP2 Technical Report WP2 D2 CATNETS EU IST FP6 003769 2006 WP4 Metrics Specification Technical Report WP4 D1 CATNETS EU IST FP6 003769 2005 WP4 Annual Report of WP4 Technical Report WP4 D2 CATNETS EU IST FP6 003769 2006 46
27. erms we assume a distributed system with N nodes where each node provides a number of resources or services There are R different types of resources which use the configuration explained in the previous section Each node knows about d other nodes called neighbours The system is modelled as directed graph G V E where each node of the graph corresponds to a node of the distributed system and there is an edge from A to each node A s neighbours Because each edge in G may not correspond to a physical link graph G is called the overlay network There is no knowledge about the size of the network We assume an overlay network where each node has d neighbors and maintains no cache about former searches When a node A needs a particular type of resource or ser vice x it always floods the network with its call for proposal messages Node A sends a message querying all or a subset of its neighbors which in turn propagate the message to their neighbors and so on To avoid overwhelming the network with search requests search is limited to a maximum number of steps t similar to the Time To Live TTL pa rameter in many network protocols In particular the search message contains a counter field initialized to t Any intermediate node that receives the message first decrements the counter by one If the counter value is not 0 the node proceeds as normal otherwise the node does not contact its neighbours A node sends only a positive response to the
28. es maturityThreshold 5 receive plumages courterThreshold 20 crossover probability crossoverProbability 0 20 mutation probability mutationProbability 0 70 ring size ringSize 100 crossOverSelectionModel 0 select plumages which are better than my plumage 1 select best received plumage crossOverSelectionModel 0 init float gene gaussWidth 0 01 min 0 001 max 0 999 CHAPTER 2 SIMULATOR REFINEMENT 23 setup genotype randomize genotype values yes no genotype randomize no if randomize no use this genotype genotype acquisitiveness 0 05 genotype satisfaction 0 99 genotype priceStep 0 5 genotype priceNext 0 05 genotype weightMemory 0 9 2 4 5 Simplified Catallactic strategy The catallactic strategy uses bilateral bargaining which comes along with a high number of messages to be transferred between the trading partners To reduce this number of mes sages a simplified version of the Catallactic strategy was implemented This simplified strategy uses the same discovery mechanism The ranking of received proposals is done according to its price Instead of starting a bilateral bargaining with the cheapest proposal the simplified strategy accepts the proposal if this proposal is lower than his budget The following learning step remains the same 2 5 Metrics This section presents the final set of metrics implemented in the si
29. esource allocator solver 1 Use LPSolve to solve the winner determination problem on the resource market resource allocator timelimit Time Out integer value in ms for the solver Valuation Generator Parameters The following parameters affect the behavior of the valuation generator valuation imitateStrategy See Section 2 3 for details valuation smallestvalue Defines the smallest value for a valuation or reservation price try to avoid zero values e valuation imitateStrategy 0 Use a normal distribution for generating valua tions and reservation prices e valuation imitateStrategy 1 Imitate the decentral strategy CHAPTER 3 GUIDE TO CONDUCTING SIMULATIONS 40 e if valuation imitateStrategy 0 valuation normal mean Mean of the normal distribution valuation normal deviation Standard deviation of the normal distribution e if valuation imitateStrategy 1 valuation strategy markepriceweight Weight of the current market price valuation strategy coldstartvaluation Value that is returned if there are no market prices or valuations in the queue cold start problem valuation strategy normalmutationmean Mean of the normal distribution to imitate mutation valuation strategy normalmutationdeviation Standard deviation of the normal distribution to imitate mutation valuation strategy normalpricestepmean Mean for the price step distribu tion valuation strategy normalpricestepdeviation Standard deviation for the price step distribu
30. for changing their deal range This feature is provided by the dynamic policy This policy uses the estimated market price of the agent to create the new deal range The estimated market price is median of the deal range its lower and upper bounds are computed using the priceNext parameter Finally the ringSize parameter specifies how many prices of successful negotiations an agent should store before he forgets these achievements Low values lead to high jumps of the markets prices whereas high values lead to very slow adaption to new market prices This file configures the catallactic strategy modify deal range values fixed static dynamic dealRange dynamic ring size ringSize 40 e learning conf This file initializes the genotype of the strategy and sets param eters of the co evolutionary learning algorithm The maturityThreshold and the courterThreshold properties control the sending and receiving of plumages In the example setting below an agent waits 5 successful negotiations until he broadcasts his plumage whereas he would wait for 20 incoming plumages until he selects one for crossover The crossoverProbability defines the probability for a gene to be selected for crossover The CATNETS strategy has 5 genes acquisitiveness sat isfaction priceStep priceNext and weightMemory If the crossover parameter is set to 0 20 one gene is chosen on average for crossover in every selection round A mutation step follows t
31. gy aln bs file The configuration file to describe the basic services aln arb file The configuration file to describe the available resource bundles cs configuration file The configuration file to describe the complex services number complexservices The number of CSs submitted during the simulation run users Determines the pattern in which ALN users submit CSs to the Complex Service Dispatcher Options 1 Simple submit CSs at regular intervals until all CSs have been submitted The interval is set by the cs delay parameter below 2 Random submit CSs at intervals which are uniformly random between zero and twice the cs delay policy Determines the scheduling policy of the Complex Service Dispatcher Options 1 Random CSs are scheduled randomly to any CSA that will run the CS 2 Queue Length schedules to the CSA with the shortest queue of waiting CSs If two CSAs have the same shortest queue length one of them is chosen at random cs delay The basic time interval in milliseconds between CSs being submitted to the ALN by the Users during simulation The actual submission interval depends on the type of user chosen see above access pattern generator Determines the order in which BSs are accessed within a CS Options 1 SequentialAccessGenerator CSs are accessed in the order stated in the CS configuration file CHAPTER 3 GUIDE TO CONDUCTING SIMULATIONS 38 2 RandomAccessGenerator CSs are accessed
32. he crossover and is performed after every successful ne gotiation The property mutationProbability defines the probability for a gene to CHAPTER 2 SIMULATOR REFINEMENT 22 be mutated The ringSize parameters stores the agreement prices of an agent in an ring array which is used for market price computation There are two policies implemented concerning how to select a genotype for crossover The first policy se lects one plumage which is better than my own current plumage whereas the second policy selects the best received plumage for crossover The property crossOverSe lectionModel sets this selection policy In CATNETS all genes are float genes with values between 0 1 To prevent the genes to reach 0 or 1 there is the possibility to set the maximum value the genes are allowed to reach with the properties min and max An important value is the size of the mutation step This size is set by the gauss Width parameter A random value is drawn from a gaussian distribution with the given width and added to the selected gene In the example the gaussWidth parameter is set to 0 01 This results in small changes of the genotypes The last section of the configuration file controls the initialisation of the genotype There is the possibility to start the simulation environment with a randomized genotype or a fixed genotype for all agents For a detailed explanation of the genotype we refer to Deliverable D2 2 setup learning send plumag
33. he same time In CAT NETS we abstracted from internal behaviour of the agents using random number gener ators We adapted here a common process used in simulation as a research method For CHAPTER 4 CONCLUSIONS 45 commercial use a more detailed decision model has to implemented which fits exactly to the given environment The advanced Grid time model of the CATNETS simulator is limited to the catallactic allocation approach This leads to long running simulations of scenarios of the central allocation approach compared to the catallactic case More person months than originally planned were needed to fix bugs of the simulator code Therefore this feature could not be implemented The same reason holds for the support of co allocation and the shared resource model Both are supported only with limitations in the catallactic model Large simulations consume lots of memory Therefore these simulations should be executed only on servers with at least 2GB of main memory 4 3 Future Extensions The simulations environment will be used and extended in the EU project eRep Repu tation will be added to the simulation scenario The implementation model of the agent will be replaced by BDI agents and electronic institutions will be introduced For more information the reader is referred to the eRep web site http megatron ilia csic es eRep Furthermore the failure model if the sites resources and service will be extended and risk managem
34. ies performed by WP2 in the third and final year of CATNETS According to the work plan of the project in year 3 WP2 was supposed to work on task Simulation of Application Layer Networks and refinement In this document the focus is on refinement i e on describing how the CATNETS simulator has been improved after the last review meeting Detailed account on performed extensive ALN simulations CHAPTER 1 INTRODUCTION 4 scenario parameters Scenario Generator scenario Simulator _ Alloc technical metrics Evaluator ecanomical performance indicator Figure 1 1 Architecture for evaluation system and simulations of prototype like scenarios is given in Deliverable D4 3 WP405 which includes a comprehensive report of experiments conducted to evaluate the catallactic and centralised allocation mechanisms Some directions for simulator refinement were identified and listed in Deliverable D2 2 WP206 Others derived from reviewers comments and suggestions included in the CATNETS Review Report N 2 In summary in project year 3 WP2 worked on the following refinement tasks Automated scenario generator This tool was needed to produce large scale scenarios to be simulated Its implementation started in project year two and was concluded as planned at 70 28 Message delivery failure The simulator was extended by adding the possibility of mod eling message delivery failure so that experiments could cover a
35. inquiring node if x is found and has enough free capacity When the search ends the inquiring node A will either have the contact information for resource x or nothing if all resources are used or currently down due to a failure In the latter case node A assumes that a node offering the resource cannot be found The implemented search strategy is pure flooding With pure flooding a node A that searches for a resource x checks its local resource and contacts all its neighbours In turn A s neighbours check their local resources and propagate the search message to all their neighbours The procedure ends when either the resource is found or a maximum of t steps is reached The scheme in essence broadcasts the inquiring message In the CATNETS simulator the number of hops is controlled by two parameters be cause of different broadcast messages e Call For Proposal and Announce message hop count The Call For Proposal mes CHAPTER 2 SIMULATOR REFINEMENT 20 sage broadcast the demand to available seller within the given hop limit An An nounce message informs all agents about the results of the discovery phase The behaviour is the following A Call For Proposal message receiver blocks its basic service instance or resource bundle until an announce message is received If the number of hops is high the requestor of e g a basic service gets a higher number of basic services for selection but blocks these basic service for any other reque
36. lso more realistic scenarios were communication between ALN sites is error prone This task was CHAPTER 1 INTRODUCTION 5 completed by 70 28 Link load dependent message latency Variable message latency depending on band width currently available between ALN sites was taken into consideration A mech anism implementing this feature already embedded in the base simulator Optor Sim was analysed adapted and included in the CATNETS simulator This task was completed as planned by T70 30 Event driven time model The base simulator OptorSim offered two time models one time based and one event driven In the last year of the project the event driven time model was adopted and embedded into the CATNETS simulator This task ended at T0 32 Refinement of centralised mechanism Some modifications to the implementation were done in order to increase the service resource allocation rate when the cen tralised allocation mechanism is adopted This task was completed by 70 28 Refinement of catallactic mechanism This tasks involved some bug fixing and the im plementation of resource co allocation It has been completed by 70 32 Full implementation of metrics At the end of project year 2 the CATNETS simulator implemented the recording of a subset of the technical metrics defined by WP4 in Deliverable D4 1 WP405 The implementation of the metrics framework was completed at TO 31 The issues listed above are described in Chapter 2 In pa
37. mulation environment The metrics are measured during simulation runs and stored in several text files Analysis scripts use these text files and compute the metrics pyramid In detail the files are e Filename accepts txt Description This file records the successfully ended negotiations accept for one simulation run Each accept equals one entry in the accepts file Sample entries 3725 CSA5 Site20 BSA0 Site24 accepts Negotiation0 seller 6534 CSA0 Site3 BSAOQSite5 accepts Negotiation2 buyer Column description time stamp in milliseconds long ID of the agent String ID of the agent String name of metric String negotiation id String role String buyer or seller which made the acceptance decision CHAPTER 2 SIMULATOR REFINEMENT 24 e Filename basic_service_allocation_time txt Description This file lists the start and end events of basic service allocations Both events are recorded by a complex service agent A complex service agent measures the start event when he issues a new basic service request When a complex service agent receives an end of negotiation signal he adds the end event to the basic service allocation time file Using these start and end events the basic service allocation time is computed as the different between the start and end event time Sample entries 3000 CSA5 Site20 basic_service_allocation_time start Negotiation0 3740 CSA5 Site20 basic_service_allocation_time end Negotiation0 success
38. n Section 2 1 3 1 4 CATNETS simulator Installing 1 Start by unpacking the code from the ZIP file by using an archiving package such as WinZip do not open the file directly from the browser but save it to disk first and then open it 2 Go down into the catnet s sim directory catnets sim is now ready to run Running e The main executable is called catnets sim bat and can be found in the main catnets simulator directory e When running catnets sim the user must have write permission to the current directory so catnets sim can write output files e catnets sim is run from the command line and takes either zero or one arguments The optional argument is the parameters file explained in the next section and if no file is specified the default parameters file located at examples parameters_catnets conf is used Usage S catnets sim bat parameters file CHAPTER 3 GUIDE TO CONDUCTING SIMULATIONS 37 3 2 CATNETS simulator parameter file The parameter file for the CATNETS simulator had already been partially described in Deliverable D2 2 WP206 For the sake of clarity the full description of the parameter file is given here again The simulation parameters are set manually by the user in a parameters file The default parameters file is examples parameters_catnets conf Following is an explanation of each parameter 3 2 1 General Parameters aln topology file The configuration file to describe the ALN topolo
39. nfigured with its initial trading price range setting a minimum and a maximum price minPrice and maxPrice The hard lower and upper limits are additional constraints preventing usuary bids A seller will never sell his basic service below the given lower limit whereas a buyer will never submit a bid higher than his budget hard upper limit All basic services instances of the same type use the same hard lower and upper limits The next configuration parameter assigns resource bundles to the given basic service type This allows a basic service agent to choose between different resource bundles and buy them on the resource market The order of the resource bundles is important a bundle at first place has higher preference than a bundle at second place If there is a list resource bundles the co allocation switch of the simulator has to be turned on This allows the basic service agent to buy from more than one resource provider The basic service agent tries to fulfill his CHAPTER 2 SIMULATOR REFINEMENT 14 resource demand with the resource bundle specified at the first place before he selects a resource bundle of the second specified type resource itemids lt BS gt seller minPrice lt double gt lt BS gt seller maxPrice lt double gt lt BS gt buyer minPrice lt double gt lt BS gt buyer max Price lt double gt lt BS gt hard lower limit lt double gt lt BS gt hard upper limit lt double g
40. nt name of the metric event type start or end negotiation id result of the negotiation only written for end events e Filename resource_buyer_bid_central txt Description Bid of a basic service agent on the resource market Sample entry 36533 Sitel4 BSAO0 Site3 resource _buyer_bid_central 13 169972617048721 rl Column description time stamp receiver site auctioneer bidding agent name of the metric valuation highest price to pay resource id Note Only used in the central case CHAPTER 2 SIMULATOR REFINEMENT 29 e Filename resource_seller_bid_central txt Description Bid of a resource service agent on the resource market Sample entry 11126 Sitel4 RAO Sitel5 resource seller bid central 3 0790178353823725 rl Column description time stamp receiver site auctioneer bidding agent name of the metric reservation price minimum price to trade resource 1d Note Only used in the central case e Filename resource_usage txt Description The resource usage one line per Sample entry 996 999 start and end event 178488804294 RAO Site28 resource_usage start Negotiation2 rl 1 178488805282 RAO Site28 resource_usage end Negotiation2 rl 1 Column description time stamp name of the writing agent name of the metric event type start or end negotiation id the resource bundle which was used e Filename service_buyer_bid_cen
41. ntify the required features of the CATNETS simulator to develop it and to incorporate the centralised and catallactic service resource market specifications from WP1 Theoretical and Computa tional Basis The simulator is to be used for controlled executions of Application Layer Network ALN scenarios Results of executions are the basis of performance evaluation and comparison of the catallactic and centralised allocation mechanisms performed by WP4 Evaluation The simulator is a component of the high level workflow depicted in Figure 1 1 which measures technical and economic metrics which the evaluation component uses for com putation of several indicators The components acting in the workflow are Scenario Generator This component takes a set of scenario parameters as an input and produces a scenario to be simulated as an output Simulator It takes a scenario as an input and executes it by using a service and resource allocation approach The output of a simulation is a set of technical and economic metrics as described in Deliverable D4 1 WP405 Evaluator This component takes a set of technical and economic metrics as input and as described in Deliverable D4 1 WP405 calculates an economical performance indicator for the allocation mechanism under observation The description of the components above is mainly given in Deliverables D2 2 WP206 and D4 2 WP406 produced in project year 2 This deliverable focuses on the activit
42. of ARBs ARB ARBs will be created each of which including a number of resources defined by a random integer in the range 1 MazRes Quality The user can add and delete values for the quality of basic services Basic Service The user can define B S the number of basic service types Complex Service The user can define C S the number of complex services and MaxBS the maximun number of Basic services per CS C S complex ser vices will be randomly created each of which including at most MazBS basic services Failure probability The user can define a range for the site failure probability Every site will fail with a probability in the range Min Fail Prop Max Fail Prop CHAPTER 2 SIMULATOR REFINEMENT 8 Boston University Representative Internet Topology Generator BRITE Topology Type ha Level AS ONLY e fas Router Top Dawn Bottom Up Catnetst Cath f Resource Parameters ARB Parameters Res ilaxRes Max ARB Quality Basic Service platinum ly ES Add Delete Complex Service 05 Max FailProb MaxBS E Min FailProb Export Topology Location Browse File Formats lv BRITE OTTER Build Topology Figure 2 2 Automated scenario generator Catnets1 parameter tab Catnets2 Tab This tab allows the configuration of the complex service basic service and resource in stances Distributions are available which assign the inst
43. otifies other MessageTransfer instances that it is starting and therefore changing the network situation and waits until the transfer is complete If the waiting is interrupted by another MessageTransfer start the time left to transfer the message is recalculated based on the new network load The MessageTransfer instance which started before waits again until the transfer is complete When the transfer has finished it notifies all the other currently running MessageTrans fer instances 2 2 3 Event driven time model The base simulator OptorSim includes two time models to be used for simulation one time based and one event driven time model Simulation runs with both time models de livering the same end results In project year 3 the event driven time model was adopted for the CATNETS simulation scenario In the time based model the simulation proceeds in real time The simulation time equals the wall clock time to complete a simulation run by simulating all complex service P2P Mediators deal with management of all messages when a message has to be sent by an agent located on a site actual delivery also to other local agents is performed by the local P2P mediator CHAPTER 2 SIMULATOR REFINEMENT 12 requests In the event driven model the simulation time is advanced to the point when the next thread should be activated The use of the event driven model speeds up the running of the simulation considerably whereas the time base
44. r proposal messages one entry for each call for proposal message sent Sample entry CHAPTER 2 SIMULATOR REFINEMENT 26 1178488801492 CSA15 Site20 cfps_sent Negotiation0 1 Column description time stamp name of the agent name of the metric name of the negotiation id number not used Note Only used in the catallactic approach e Filename complex_service_allocation_rate txt Description Allocation rate for each complex service Sample entry 1178498801492 CSA1 Sitel complex_service_agent_allocation_rate 0 10183299389002037 Column description time stamp name of the complex service name of the metric allocation rate e Filename complex_service_provisioning_time txt Description Sum of basic service provisioning times This includes successful and failed allocations Sample entry 1178488805241 CSA10 Sitel3 complex_service_provisioning_time 2749 failure Column description time stamp name of the agent name of the metric complex service provisioning time failure success e Filename CS_BS_Mapping txt Description Mapping of complex service request to basic service negotiation id Sample entry 1178488803204 CSA10 Sitel3 CS_BS_Mapping cs1_2 Negotiationl Column description time stamp name of the agent name name of the metric id of the request related basic service negotiation id Note The current central implementation does not support this metric e Filename csa_demand_distribution txt De
45. r the CATNETS simulator are created e topology conf specifies the ALN topology e cs conf specifies the complex services e bs conf specifies the basic services e arb conf specifies the available resource bundles Two additional files are created e lt filename gt brite specifies the topology in BRITE format e lt filename gt odf optional specifies the the topology in OTTER format see http www caida org tools visualization otter lt filename gt can be specified by the user 2 2 Infrastructure 2 2 1 Message delivery failure The communication system of the CATNETS simulator developed in project year 2 fea tures both point to point and broadcast multicast communication These communication paradigms are used to implement the negotiation protocols adopted in the centralised and catallactic markets In real networks the delivery of point to point or brodcast multicast message can fail the possibility of simulating failure of ALN links has been added to the CATNETS simulator in project year 3 Given the base simulator OptorSim the failure probability was associated to ALN sites instead of links because of easier implementation 1 Every ALN site has a given failure probability f This is a static parameter spec ified in the configuration file topology conf The failure probability can be specified using either the manual or automated scenario generators CHAPTER 2 SIMULATOR REFINEMENT 11 2 The si
46. rticular Section 2 1 describes the automated scenario generator while Section 2 2 presents the enhancements to the simula tor s messaging system and time model Sections 2 3 and 2 4 show the implementation refinements of the centralised and catallactic markets Finally Section 2 5 gives an ac count on the full implementation of technical metrics and their semantics In the rest of the document Chapter 3 is a user guide which explains how to conduct simulations and Chapter 4 summarizes the achieved results and gives directions about how the CATNETS simulator can be further extended Chapter 2 Simulator refinement 2 1 Automated scenario generator The CATNETS automated scenario generator is a tool used for the generation of scenar ios to be given as an input to the CATNETS simulator In Deliverable D2 2 WP206 the requirement for this tool were identified In this section the focus lies on its implementa tion The automated generator is based on BRITE MLMBO1b a topology generation framework which is able to generate synthetic topologies that accurately reflect many as pects of the actual Internet topology BRITE supports multiple generation models includ ing models for flat AS flat Router and hierarchical topologies Models can be enhanced by assigning links with attributes such as bandwidth and delay Figure 2 1 shows how to choose the topology model in BRITE For details about how to use BRITE the reader is referred to the BRI
47. s the creation of the market product configuration file Improvements to the strategy are presented in the following subsections In subsection 2 4 3 a performance analysis of the flooding algorithm is introduced Several simulation runs evidenced that the performance of the discovery algorithm has large impact on the allocation performance in the Catal lactic allocation approach The performance analysis gives advice how to set the number of hops for the flooding algorithm A simplified Catallactic strategy implementation is presented in subsection 2 4 5 In stead of using a iterative bargaining with several negotiation rounds this strategy uses only one negotiation round for reaching an agreement This helps to reduce the messag ing complexity of the Catallactic iterative bargaining implementation 2 4 1 Configuration of the Catallactic market The configuration of the Catallactic market includes the initialisation of the traded prod ucts its price ranges and several constraints of the Catallactic market These constraints improve the performance of the Catallactic strategy in the examined scenarios A detailed description of the Catallactic market configuration contains the following properties e Service market products Each basic service type BS which is traded on the service market has to be configured for the Catallactic strategy The complex service agent as a buyer and the basic service agent as a seller of the service can be co
48. scription The distribution of the complex service demand Sample entry y 178488801492 CSA15 Site20 csa_ demand _ distribution csl cs1_ 1 178488802492 CSA10 Sitel3 csa_demand_distribution csl csl_2 CHAPTER 2 SIMULATOR REFINEMENT 27 Column description time stamp name of the agent name of the metric type of the requested complex service id of the complex service request e Filename distance txt Description The distance between the seller and buyer Sample entry 1178488803219 CSA10 Sitel3 BSA0 Site23 distance 2 Negotiationl Column description time stamp name of the buyer agent name of the seller agent name of the metric distance negotiation id e Filename latency txt Description Latency between two negotiation partners Sample entry 1178488804211 CSA14 Sitel7 BSA0 Site28 latency 31 Negotiation3 Column description time stamp name of the buyer agent name of the seller agent metric name latency negotiation id e Filename market price resource central txt Description Market prices for resources on the resource market Sample entry 1178748042878 RMAAO CentralAuctioneerSite market_price_resource_central rl 7 692018032073975 Column description time stamp site name of the metric resource type price Note Only used in the central case e Filename market_price_service_central txt Description Market prices for services on the service market Sample entr
49. sts un til the announce message is received This blocking policy leads to a low allocation rate of the system for a high number of hops defined in the hop count parameter and produces a high number of messages which have to be sent over the P2P network Addressing both problems the number of hops should be kept fairly low to 2 or 3 hops for a network with a high density of available services or resources If there is a low density of available services this number should be increased To overcome the problems of the implemented blocking policy the discovery time of a requestor the time to receive an answer for a call for proposal message is randomized This gives the seller agents the opportunity to respond to other call for proposals when ever they receive a reject with an announce message for the current selection If the requestor would like to execute a long running service there is the possibility to specify a counter of how many discovery time periods a requestor should wait receiving an answer for his call for proposal message Of course this increases the overhead time until the service execution can start In the example implementation below a requestor decides to wait 10 times a ran domized discovery timeout until he gives up his search for a basic service instance or a resource bundle Random rand new Random _time gtSleep int _params getDiscoveryTimeout Math round int _params getDiscoveryTimeout rand nex
50. t lt BS gt resource itemids lt RB gt lt RB gt An example configuration is presented below Both the complex service agents and the basic service agents start with the same initial price range The budget of the complex service is 50 the minimum costs for the basic service bs1 is 20 This lower bound can be seen as the production costs for a basic service provider Both the seller and buyer start with the same price range between 25 and 40 The specified basic type is only allowed to buy resources of product r1 bsl seller minPrice 25 bsl seller maxPrice 40 bsl buyer minPrice 25 bsl buyer maxPrice 40 bsl hard lower limit 20 bsl hard upper limit 50 bsl resource itemids rl e Resource market products The reason for specifying resource market products is the general demand of resources in the basic service configuration The strategy works on product level This means the catallactic reasoner estimates his price for a given set of traded products For a resource bundle of the basic service config uration either each single resource is traded and its prices are summed up or a predefined set of resource bundles can be traded on the resource market The Catal lactic mechanism chooses the second option which allows a more flexible way to trade products on the resource market We assume there are base resources on the market Each resource provider defines his individual set of resource
51. t lt resource_product_id gt seller max Price lt double gt lt resource_product_id gt buyer minPrice lt double gt lt resource_product_id gt buyer max Price lt double gt lt resource_product_id gt hard lower limit lt double gt lt resource product id gt hard upper limit lt double gt arb itemids lt resource roduct d gt lt resource roduct d gt An example of one resource product is shown below The resource bundle consists of three base resources Multiples of 1 ram unit 2 cpu units and storage unit are tradeable Both resource buyer and resource seller start with the same initial price range The budget of the buyer is set to 50 and the seller s minimum price is 20 There is only one product tradeable on the resource market The resource providers check if they are able to provide resource bundle r1 rl resourceids ram cpu storage rl baseunit ram 1 rl baseunit cpu 2 rl baseunit storage 1 rl seller minPrice 25 rl seller maxPrice 40 rl buyer minPrice 25 rl buyer maxPrice 40 rl hard lower limit 20 rl hard upper limit 50 arb itemids rl CHAPTER 2 SIMULATOR REFINEMENT 16 2 4 2 Market Property Generator A visual tool the market property generator MPG was developed to help creating this property file for the catallactic market configuration The market property generator reads the arb conf and bs conf files which are created
52. t catallactic implementation because of its high complexity In the catallac tic approach we assume the co allocated resources have the same resource bundle size capacity and product id The central approach supports all co allocation combinations which allows allocation of bundles with different size and capacity 43 CHAPTER 4 CONCLUSIONS 44 The service market decouples the service requests from the resource market A com plex service doesn t have to know how many resources there are and how many resources he needs for his service The complex service can focus on creating value added services to the user Currently the complex service sequentially request a list of basic service This could be enhanced in future releases of the CATNETS simulator with more sophisticated workflow engines As on the resource market an service allocation policy is applied to al locate services In CATNETS we implemented two allocation approaches a continuous double auction and the catallactic bargaining Supporting both allocation mechanisms the simulator provides proactive and reactive interfaces for software agents Proactive agents act on their own They periodically check if there is new demand or supply and send their bids to the auctioneer In the reactive agent model the agents wait for new events like incoming messages before they act on new Situations The simulator provides a rich messaging model including a large set of different mes sage
53. t save it to disk first and then open it 2 Go down into the catnet s manual scenario generator directory catnets manual scenario generator is now ready to run Running e The main executable is called catnets msg bat and can be found in the main catnets manual scenario generator directory e When running catnets msg the user must have write permission to the current directory so catnets msg can write output files e catnets msg is run from the command line and takes zero arguments Usage catnets msg bat e The usage of the manual scenario generator is explained in D2 1 WP206 Section 3 3 1 3 CATNETS automated scenario generator Installing 1 Start by unpacking the code from the ZIP file by using an archiving package such as WinZip do not open the file directly from the browser but save it to disk first and then open it 2 Go down into the catnets automated scenario generator directory catnets automated scenario generator is now ready to run CHAPTER 3 GUIDE TO CONDUCTING SIMULATIONS 36 Running e The main executable is called catnets asg bat and can be found in the main catnets manual scenario generator directory e When running catnets asg the user must have write permission to the current directory so catnets asg can write output files e Catnets asg is run from the command line and takes zero arguments Usage catnets msg bat e The usage of the automated scenario generator is explained i
54. tDouble while _proposals isEmpty amp amp discoveryCounter lt 10 _time gtSleep int _params getDiscoveryTimeout Math round int _params getDiscoveryTimeout rand nextDouble discoveryCounter e Learning message hop count This hop counter controls the speed of spreading plumages to other agents of the same role The number of hops should also kept low because of the increasing number of messages which are needed to flood the network with new plumage information CHAPTER 2 SIMULATOR REFINEMENT 21 2 4 4 Configuration of the catallactic strategy and learning algo rithm In year 3 of the project the parameters of the catallactic strategy and the learning algorithm were externalized to ease the configuration for simulation runs Two configuration files were introduced e strategy conf This file contains a the configuration of the catallactic strategy The dealRange property defines which policy to use for adapting the deal range after a negotiation The value fixed leads to not changing the deal range after a negotiation The price ranges of the market configuration stay the same in every negotiation during the simulation run The static policy adapts the price range according to the priceNext parameter of the genotype The length of the deal range stays the same as defined in the market configuration The fixed and static policies don t take into account the market price estimation
55. te failure probability f is used by P2P mediators located in every site when ever a P2P mediator is requested to delivery a message the message is forwarded to the recipient s with a probability p 1 fp If the failed message is a point to point message a single agent will not receive it In other words the link between one sender and one receiver agent is not working If the message is a multicast broadcast message the P2P will not propagate it to neighbour sites and multiple agents will not receive it In this case the site failure probability can be interpreted as the failure of all site s out links This implementation of message delivery failure permits the simulation of a wide range of real situations including the real world characteristic that failure of broad cast multicast delivery has certainly a greater overall impact on the course of simulation than point to point delivery failure 2 2 2 Link load dependent message latency The CATNETS simulator has been refined by adding the possibility of simulating deliv ery of messages dependent on the current network traffic 1 e how much bandwidth is consumed by the messages currently being delivered along the path between sender and receiver The message transfer implementation provided by OptorSim was adapted Every mes sage transfer between sites is handled by an instance of the class MessageTransfer This class takes care of message transfers over the simulated ALN It n
56. time site name of metric event type start end time e Filename successful_CS_request txt Description Successful complex service requests one line for each complex ser vice Sample entry 178498801492 CSA4 Site7 successful_CS_requests 2 178498801492 CSA0 SiteO successful_CS_requests 21 Column description time stamp name of the writing agent name of the metric successful cs requests e Filename total_cs_requests txt Description Total complex service requests received by each complex service one line for each cs Sample entry 1178498801492 CSA4 Site7 total_CS_requests 465 1178498801492 CSAO0 Sited total_CS_requests 507 Column description time stamp name of the writing agent name of the metric number of cs requests e Filename trade_resource_central txt Description Maps basic service agents and resource service agents that trade Sample entry CHAPTER 2 SIMULATOR REFINEMENT 31 1179784806500 Negotiation29 trade _resource_central Negotiation0 Column description time stamp negotiation Id buyer basic service name of the metric negotiation Id seller resource service e Filename trade_service_central txt Description Maps complex service agents and basic service agents that trade Sample entry 1179784817625 Negotiation8l trade _service_central Negotiation88 Column description time stamp negotiation Id buyer complex service name of the metric
57. tion valuation strategy depthweightedaverage Depth of the weighted average maximum age of the historical price information valuation strategy buyersellermultiplier Scaling factor for the generated valuations and reservation prices 3 2 3 Market Parameters The following parameters are specific for the service resource allocation mechanisms market model Set to 1 to use the catallactic allocation mechanism or 2 to use the centralised mechanism market decentral file Configuration file including parameters for the catallactic market see Section for detatils price range randomize Randomize initial price range of the agents This parameters gives the possibility to randomize the initial price ranges of the catallactic market configuration A value of 0 means change of the initial price range market connect Connect the prices of the service market and resource market If value is yes the basic service seller s outcome is the budget of the basic service buyer on the resource market resource model Resource model selection resource values shared dedicated The catallactic strategy implementation is able to handle dif ferent resource models In the shared resource model the resource provider can CHAPTER 3 GUIDE TO CONDUCTING SIMULATIONS 41 allocate different resource bundles to different basic services The dedicated resource model allocates the whole resources exclusively to one basic service cfp selection mod
58. tral txt Description Bid of a complex service agent on the service market Sample entry 17615 Sitel4 CSA7 Site23 service_buyer_bid_central 8 995795995659913 bsl Column description time stamp receiver site auctioneer bidding agent name of the metric valuation highest price to pay service id Note Only used in the central case e Filename service_seller_bid_central txt Description Bid of a basic service agent on the service market Sample entry 11136 Sitel4 BSA0O Site29 service _ seller bid central 14 11408399533697 bsl Column description time stamp receiver site auctioneer bidding agent name of the metric reservation price minimum price to trade service id Note Only used in the central case CHAPTER 2 SIMULATOR REFINEMENT 30 e Filename service_usage txt Description The service usage one line for each start and end event Sample entry 178488804195 CSA10 Sitel3 BSA0 Site23 service_usage start Negotiationl 1 178488805256 CSA10QSitel3 BSA0 Site23 service_usage end Negotiationl Column description time stamp name of the buyer name of the seller name of the metric event type Start or end negotiation id an integer value deprecated e Filename simulation time txt Description The time needed for simulation Sample entry O NoSite simulation_time start 0 10003325 NoSite simulation_time end 10003325 Column description
59. y 1178748022048 SMAAOf CentralAuctioneerSite market_price_service_central bs1 4 70427463424487 Column description time stamp site name of the metric service type price Note Only used in the central case e Filename negotiation_messages txt Description Negotiation messages sent for an accept or reject Sample entry CHAPTER 2 SIMULATOR REFINEMENT 28 178488802297 BSA0O Site9 negotiation_messages servic reject Negotiation0 15 178488803219 BSA0 Site23 negotiation_messages servic accept Negotiationl 5 Column description time stamp name of the agent which write the metric name of metric market result of negotiation negotiation id number of messages Note Only used in the catallactic approach e Filename rejects txt Description List of rejects and the role which rejected Sample entry 1178488802297 CSA15 Site20 BSA0D Site9 rejects Negotiationo buyer Column description time stamp name of the buyer agent name of the seller agent name of the metric negotiation id role which rejected Note Only used in the catallactic approach e Filename resource_allocation_time txt Description The resource provisioning time includes the execution time Sample entry y 178488803219 BSA0 Site23 resource_allocation_time start Negotiation0 178488804195 BSA0 Site23 resource_allocation_ time end Negotiation0 success Column description time stamp name of the writing age

Download Pdf Manuals

Related Search

Related Contents

保証規定：PDF/185KB  Samsung SH-B083L  DENVER CR-215 8-languages  Mode d`emploi Konftel 300    Sonic Blue Rio EX1000 User's Manual  Profiler™ 2  

Copyright © All rights reserved. Failed to retrieve file