Home

Thesis Project - Simple search

1. last min avg max W incoming traffic on interface ethO avg 1 2KBps SOSBps 1 88MBps 14 53 MBps W incoming traffic on interface ethl no data Figure 20 LTSP Test 3 Network 192 168 186 111 ping statistics 380 packets transmitted 380 received 0 packet loss time 379108ms rtt min avg max mdev 0 234 2 611 22 357 4 282 ms The ping test from thin client2 show in the topology see appendix A showed that the latency became slightly higher but there were no loss of packets Appendix D Openthinclient server baseline The following graphs show the baseline of the Openthinclient server otc server CPU Load 1h 0 1 005 RS a j E peed fl ple AA e A RRERBESTISILSSLLRRARRRARRRASSSSSSSSSESLSSNRRSEHERAARTSARSERERSRRSSRERE EB 3 di 3 last min a max E Processor load avg o o 001 0 07 I Processor load15 avg 0 01 0 0033 0 05 0 27 E Processor load5 avg 0 0033 0 001 0 03 Figure 21 Openthinclient Baseline CPU Load Figure 20 shows the processor load for one minute displayed in green processor load over 5 minutes displayed in red and processor load over 15 minutes displayed in blue The horizontal axis displays time in minutes while the vertical axis shows the load of the processor If the processor load reaches 1 it means that one core of the CPU has reached 100 load otc server CPU Load 2 1h 3 a last
2. Second test Itsp CPU Load 1h last min avg max W Processor load15 avg 0 18 0 07 018 034 W Processor load5 avg 0 31 0 03 O21 058 H Processor load avg 0 3 0 0 O Trigger Processor load is too high on Itsp gt 5 Figure 17 LTSP Test 2 CPU Itsp Network 1h 9 54 MBps 7 63 MBps 5 72 MBps 3 81 MBps 191 MBps OBps 282738232252238923232353233323922393353293333333 5 Senet etkhebpesescseetees 3 last min avg max W incoming traffic on interface ethO avg LO6KBps SOSBps 674 99KBps 8 89MBps E incoming traffic on interface eth no data Figure 18 LTSP Test 2 Network The second test also showed unaffected memory and disk space so those graphs will not be shown The ping results showed that the latency became slightly higher but there was no loss of packets Below follows the final test configured to simulate 15 users Since memory and disk space still showed to be unaffected these will not be shown Third test Itsp CPU Load 1h SGORRRRRRSSSSSSSSSSRSBARRSEARRRRRRARRSKRRRRRRRARRARRKRRRSSIIIS RISES S a last min avg max W Processor load15 avg 013 013 036 076 Processor loads avg 0 07 003 04 1 09 E Processor load favg 0 01 0 039 154 Trigger Processor load is too high on Itsp gt 5 Figure 19 LTSP Test 3 CPU Itsp Network 1h 15 26 MBps 13 35 MBps 11 44 MBps 9 54 MBps 7 63 MBps 5 72 MBps 3 81 MBps 191 MBps OBps
3. 20 Itsp CPU Load 1h ax W Processor load15 avg 018 007 018 0 34 W Processor load5 avg 0 31 003 021 058 H Processor load avg 0 38 0 024 O Trigger Processor load is too high on Itsp gt 5 Figure 4 LTSP server test 2 CPU performance When comparing the measurement from the first performance test the results show that the load on the processor has increased slightly The graph shows an increasing irregularity in the processor load and the baseline becomes hard to read out The final performance measurement of the LTSP solution is conducted by increasing the number of simulated users to 15 and by using the original script after modifying the send yml file to increase the bandwidth and packet_size attributes ten times This test has shown the following CPU Load Itsp CPU Load 1h 16 14 12 avi W Processor load15 avg 013 013 036 07 Processor loads avg 007 003 04 1 E Processor load favg 0 01 0 039 15 O Trigger Processor load is too high on Itsp gt 5 Figure 5 LTSP server test 3 CPU performance When comparing the measurement from the baseline the results show a large increase of the processor load The graph shows that the load from the background process is quite low in comparison to the load generated from the test The baseline is hard to read out in this graph which indicates that the system is having some difficulties returning to the normal behavior
4. 2011 Available from http www knowledgeonecorp com news pdfs Thin 20client 20vs 20Fat 20client 20Computing pdf Reese G 2000 Distributed Application Architecture In Database Programming with JDBC and Java 2nd ed online Accessed 23rd Jan 2011 Available from http java sun com developer Books jdbc ch07 pdf O Reilly Media Rizwan A 2011 Upgrading and Performance Analysis of Thin Clients in Server Based Scientific Computing online Accessed 20th April 2011 Available from http liu diva portal org smash get diva2 400152 FULLTEXTO02 Schmidt B K Monica S Lam J and Northcutt D 1999 The interactive performance of SLIM a stateless thin client architecture online Accessed 23rd Jan 2011 Available from http labs oracle com techrep Perspectives PS 01 5 pdf search 22thin client 20history 22 Tykesson A Ericsson T 1999 Tunna eller tjocka klientapplikationer en studie i Client Server teknik online Accessed 20th April 2011 Available from http gupea ub gu se handle 2077 1360 Tyson B 2010 The History of Thin Clients online Accessed 27th Jan 2011 Available from http www brighthub com environment green computing articles 71173 aspx Homepage of Zabbix An Enterprise Class Open Source Distributed Monitoring Solution online 2011 Accessed 20th April 2011 Available from http www zabbix com Openthinclient online 2011 Accessed 20th April 2011 Available fro
5. computer which does not do any local computation but instead sends the information to a centralized server where the operation will be performed 2 4 2 Product solutions There are many different products solutions available regarding thin client environments Choosing which one of these to pick might be hard because the companies distributing these might not give a fair image as to how good or bad their respective solution is In order to narrow the selection down this thesis will only compare two different product solutions In this case the solutions that will be compared are Openthinclient and LTSP Linux Terminal Server Project While there are many available product solutions for thin client environments both commercial and open source the chosen products are both open source products The reason why two open source products were chosen instead of more commercial products like Sun ray and Citrix is due to the fact that their solutions have a high cost and without proper funds open source is the only alternative 2 4 3 Compared products The explanation of product solutions is fetched directly through the given information of the products from the respective company distributing them There are many available options when choosing among the thin client products However most of them do not have available support other than forums and mail Due to this the chosen products are LTSP and Openthinclient LTSP Linux Terminal Server
6. min ag max M Number of processes avg 29 28 28 7 29 M Number of threads vi 447 440 444 3 451 gl M Number of running processes no data Figure 22 Openthinclient Baseline CPU Threads and Processes Figure 21 shows the number of processes displayed in green number of threads displayed in red and number of running processes displayed in blue The horizontal axis displays time in minutes while the vertical axis shows the amount of threads otc server Harddrive 1h R RARRRRSSSSSSSSSSLSAARAASERREARARRAKHRAERASSRRRRRRERRFSTVIPCIR SSSR RAR MM last min avg max H File read bytes per second lavg OBps OBps 17 12KBps 64 53 KBps W Average disk read queue length avg o o IM File write bytes per second favgl OBps OBps 67 14Bps 7 87KBps Figure 23 Openthinclient Baseline Hard drive read and writes per second Figure 22 shows the amount of bytes read per second displayed in blue the average disk read queue length displayed in red and amount of bytes written per second displayed in green The horizontal axis displays time in minutes while the vertical axis shows the amount of bytes otc server Memory 102 GB 101 GB 1 01 GB E ese AES 1 01 GB Sa Gene mien wwe 101 GB 101 GB 101 GB aaa PELLEGR EEELLLREPLELLLLL LLLP PLP L LSE LPPLSG SSPE LLL TLR P LPR TLEEEE S 3 last min avg max E Free memory avg 1016B 1 01 GB 1 01 GB 1 01 GB Trigger Lack of free memory on server otc server lt 100
7. on Windows Installing an agent on Linux is an easy task if you have a version of Linux that supports the program apt get By using this command you install the agent with all the needed dependencies The full command is as follows apt get install zabbix agent The agent also needs to be configured to point at the Zabbix server so the system knows where to send the information This configuration is done by editing the server IP address in the following file etc zabbix zabbix_agentd conf Installing an agent on Windows can be done by using a pre compiled installation package available on the Internet but also by downloading the source file and compiling the package manually In this case the first option was chosen During the installation the user is prompted by the IP of the Zabbix server 5 4 2 Configuring Zabbix Zabbix is constructed with three major components for configuration host groups templates and hosts A host is a normal device on the network that can be monitored by Zabbix using its agents while a host group is a collection of hosts The templates are used to have certain types of monitoring on different types of systems These components should according to the user manual at the official Zabbix homepage be used as follows The hosts should be named aptly and should preferably be named in Zabbix with their DNS name They should then be placed into an appropriate host group that is suited for
8. that might affect the overall performance of a computer such as memory usage processor load network load and hard drive space The items that will be measured from these attributes are e Memory usage o Cached memory 11 To store information into a cached memory is to store the information in a place that has faster access time than the normal physical memory o Shared memory Typically it refers to a large block of memory that can be accessed by several CPU s simultaneously in a multi CPU computer o Free memory e Processor load o Average processor load one minute o Average processor load five minutes o Average processor load fifteen minutes e Network load o Incoming traffic on the network interface e Hard drive space o Size of remaining space of the partition The remaining size of the partition containing the operating system o Free disk space on home This purpose of this measurement is to see if the users home folders are affected by the thin client solutions o Free swap space The swap space is a place where information can be stored temporarily to allow a computer to read and write the data into the physical memory faster than normal on a hard drive All of the compared solutions are monitored this way with Zabbix The graphs extracted from these attributes are shown in appendix B E 4 4 Comparing prices In order to give a correct comparison between different product solutions it is imp
9. the thin client is a tool for user interaction that forwards the users request to the centralized server where the request is processed and then sent back The thin client is more or less a gateway between the user and the server to request information from a server and to display the result The client itself does not do any operations that require hardware but instead delegates it to the server This does put a larger constraint on the server but it also means that the client can work on minimal hardware Since the thin client does not need any particular hardware it is cheaper simpler and more reliable than the fat client This means that the thin client can be deployed in harsher environments 2 4 1 History of Thin Clients According to LLC Books 2010 the first computers were used as an automated calculation device There was no real time interaction as we know of it today only a string of numbers as input and output These computers slowly developed different input methods like hole punch cards paper tape magnetic tapes and eventually a kind of terminal that acted like a typewriter The research continued which led to simple VO Input Output devices which had a keyboard as the input device and a standard CRT Cathode Ray Tube display as the output device These devices only had one input device and one output device which when several users tried to interact with led to the development of a centralized node a mainframe with dumb t
10. to be added into the LTSP tree to be allowed to contact the server which can be configured in ltspadmin In order for the clients to be authenticated their DSA key needs to be copied into the servers LTSP tree After the key has been placed there the clients can start to use the LTSP tunnel This concludes the configuration of LTSP on the server however the Zabbix agent still needs to be installed For more information regarding the installation of the agent see section 5 4 1 5 6 2 Performance The first step to measuring the baseline of this configuration is to add the hosts into Zabbix and monitor the hosts for a couple of hours without any generated load or traffic to see how the system behaves normally For more information about the other measured attributes see appendix B The baseline of one hour has shown the following CPU Load 19 Itsp CPU Load 1h last min ag max Processor load15 avg 0 15 0 04 O11 0 22 Mi Processor load5 avg 0 22 002 015 0 49 E Processor load avg 0 63 o 016 1 22 Trigger Processor load is too high on Itsp gt 5 Figure 2 LTSP server baseline CPU performance The graph shows the processor load for one minute displayed in blue processor load over 5 minutes displayed in red and processor load over 15 minutes displayed in green The horizontal axis displays time in minutes while the vertical axis shows the load of the processor The baseline measurements of LTSP
11. 00 Figure 24 Openthinclient Baseline Memory Figure 23 shows the amount of free memory displayed in green The horizontal axis displays time in minutes while the vertical axis shows the amount of memory 192 168 186 88 ping statistics 174 packets transmitted 174 received 0 packet loss time 172985ms rtt min avg max mdev 0 261 0 516 28 385 2 139 ms Appendix E Openthinclient server Performance tests The following graphs show the performance graphs of the Openthinclient solution during the tests The test was conducted for fifteen minutes so the first quarter of the graphs is the relevant parts The other part of the graph is the Server returning to the normal state Below is the first test conducted with five simulated users The first test showed unaffected CPU threads and Hard drive so those graphs will not be shown First test otc server CPU Load 1h last min ag m la E Processor load avg o o 001 W Processor load15 avg 0 0089 0 0067 0 01 0 03 E Processor loadS avg o o 0 01 Figure 25 Openthinclient Test 1 CPU Load otc server Memory 1h 101 GB 101 GB eee ae E A 101 GB 101 GB 101 GB 4 last min avg ax E Free memory aval 1 01 GB 101GB 1 01GB 1 01 GB 0 Trigger Lack of free memory on server otc server lt 10000 Figure 26 Openthinclient Test 1 Memory 192 168 186 88 ping statistics
12. 12 000 EURO 1 server Annual addition price 750 EURO 1 350 EURO 5 200 EURO for a further server Annual charge per 5 EURO 8 EURO 30 EURO ThinClient up to 100 clients Annual charge per 6 EURO 20 EURO ThinClient above 100 clients Annual surcharge for 420 EURO 650 EURO 2 500 EURO support of third party hardware per hardware type Table 6 Openthinclient support These categories references to the criticality of the incident where category 1 is the highest incident 11 Support Incident Project Managed Level Response Hours 24 Hours Mon Fri 8am 5pm EST Mon Fri Time ce Fri Method Email Chat Email Chat Phone Email Chat Phone On Site Scope None Application customization Application customization Application development Application development Maintenance Maintenance OS Upgrades OS Upgrades Remote Server Access Remote Server Access Research amp Development Research amp Development Backup solution Server Monitoring SLA Table 7 LTSP support 12
13. 3 25 23 36 23 37 23 38 E Processor load E Processor load15 E Processor load5 Figure 29 last min ag max 0 avg o 134 307 lava 01 01 169 301 fava o 0 144 302 Openthinclient Test 3 CPU Load otc server Memory 1h 101 GB 101 GB 101 GB f 101 GB 101 GB 168 S 168 1GB 1022 34 MB 4 23 45 23 46 09 04 23 46 last min avg max E Free memory aval 1 01 GB 1023 55 MB 1 01 GB 1 01 GB O Trigger Lack of free memory on server otc server lt 10000 Figure 30 Openthinclient Test 4 Memory The ping test from thin client2 show in the topology see appendix A showed that the latency became slightly higher but there were no loss of packets 10 Appendix F Support costs The following table shows the different support levels with their respective costs Support level Business Enterprise Support over Web support Web support portal Web support portal portal E mail E mail telephone E mail telephone telephone remote remote hands remote hands hands Response times during Category 1 48 h Category 1 12 h Category 1 8 h service hours Category 2 4bd Category 2 48 h Category 2 24h Category 3 30 bd Category 3 14 bd Category 3 10 bd Service hours Monday to Friday Monday to Friday Monday to Sunday except weekends except weekends and including holidays and holidays holidays Annual basic rate for 1 550 EURO 2 950 EURO
14. 5 6 3 Cost This solution is an open source product that is under the license called GNU GPL which means that it does not cost anything to get access to the source code This also means that downloading and installing the thin client software does not cost anything LTSP themselves do not appear to sell preconfigured hardware appliances with their solution but there are available options from disklessworkstation com This homepage also provides with three different support levels which are called incident project support and managed support In the cost calculation the chosen support level is incident For more information see appendix F LTSP offers three different thin client hardware boxes They differ in their hardware and their cost It may be important to choose a hardware box that has a Gigabit interface and a DVI port installed to support an expanding business The chosen 21 hardware box in the cost calculation is the cheapest hardware box with a Gigabit interface which in this case is the model called LTSP Term 1620 Estimation Initial cost The initial cost will consist of buying a separate server and choosing clients from among the available hardware boxes Estimation Low to medium cost Running cost Power consumption ranged from 10 Watt to 14 Watt depending on the chosen hardware Estimation Very low cost Support cost Since there are three levels of available support the cost varies greatly depending on
15. 622 packets transmitted 622 received 0 packet loss time 621137ms rtt min avg max mdev 0 237 0 947 45 727 2 028 ms The ping test from thin client2 show in the topology see appendix A showed that the latency became higher but there were no loss of packets Below follows the second test conducted with seven simulated users Second test otc server CPU Load 1h last min avg max Processor load aval o 0 002 02 Mi Processor load15 ava 0 01 0 0022 0 02 0 06 W Processor loads fava 0 0033 O 002 0 08 Figure 27 Openthinclient Test 2 CPU Load otc server Memory 1h 101 GB 101 GB ISR SSA o sl ne bel et 3 xa AA ee 101 GB 101 GB 1 01 GB 101 6B VA 1 01 6B 101 GB 101 GB 21 45 21 46 22 19 2 last mi avg max Mfree memory avg 1 01GB 101GB 101GB 101GB Trigger Lack of free memory on server otc server lt 10000 Figure 28 Openthinclient Test 2 Memory The second test also showed unaffected CPU threads and Hard drive so those graphs will not be shown The ping results showed that the latency became slightly higher but there was no loss of packets Below follows the final test configured to simulate 15 users Since memory and disk space still showed to be unaffected these will not be shown Third test otc server CPU Load 1h 23 03 23 04 23 05 23 09 23 10 2
16. 9 301 E Processor loads avg o 0 144 302 Figure 9 Openthinclient Test 3 CPU Load When comparing the measurement from the baseline the results show a large increase of the processor load The vertical axis is on a completely different scale but the server seemed to manage the performance test The baseline is impossible to read out in this graph which indicated that the load from the background process is negligible in comparison to the load generated from the test Even though the latency during this test is higher than that of the baseline measurement the server could cope with this amount of traffic 5 7 3 Cost This solution is an open source product that goes under the license called GNU GPL v2 which means that the software is free to download and install Openthinclient offers three levels of support basic business and enterprise These levels differ greatly in support medium service hours response time during service hours and cost Depending on the size of the company and the requirements of availability on support the choice may vary In the cost calculation the chosen support level is Basic For more information see appendix F Openthinclient offers many different thin client hardware boxes They differ in their hardware and their cost It may be important to choose a hardware box that has a Gigabit interface and a DVI port installed to support an expanding business The chosen hardware box in the cost calculati
17. 9100 680 Watt Hour 900 Watt Hour Support cost Medium to high Medium to very high Table 5 Cost comparison summary Translated from U S Dollars to Euro 2011 06 03 29 7 Conclusion According to Cantoria 2010 one of the weaknesses of thin client solutions are that thin clients work poorly when the latency of the network is high This thesis does however show a network with a fairly high latency during the performance tests without any lack of functionality The results indicate that both of the solutions would be able to handle double that amount of traffic and simulated users without any problems The comparison then comes down to the functionality and cost The functionality of the LTSP solution are quite limited The options they have are limited to Linux distributions without the availability of Windows The functionality of the Openthinclient solution however are quite flexible They support Linux and Windows and they have the ability to integrate their system with an existing directory service Both of the solutions have available hardware for sale with different specification depending on the needs of the company Giving a final answer to which product is the best is a challenging task since they both have strong and weak sides Depending on what the customer is after when looking to invest into an open source alternative of thin client solutions the result will be different Openthinclient is strong when it comes to f
18. A technical comparison between several different thin client solutions would be preferable 1 2 Purpose The purpose of this report is to analyze specific attributes between several thin client environments distributed by different companies to make a detailed comparison The attributes that will be compared are e Performance o Bandwidth utilization o Response time e Requirements of the server o Memory o Processor o Secondary Storage e Costs o Initial costs o Running costs o Support costs The reason using those parameters is due to the fact that if all of them are taken into account it provides a holistic perspective when comparing different solutions The purpose of the comparison using the attributes listed above is to find out which of the selected products is best in the respective area The product seen as the best here will be the product which has the overall better result from the measurement in combination with the cost 1 3 Limitations The main scope of this report is to compare several selected thin client environments to give an objective view of the different products The structure and cost of the solution will be taken into consideration whereas the companies providing these will not The limitations are set on a technical level to the specific products and or solutions There is a limitation in the number of solutions that will be tested Only two different solutions will be tested in this case The
19. Ea EE ea S 14 IE tal me Zabbix ciie ss ara a a a a e a deg teases 14 8 LLANA id 16 o AS A A res nao lossna i r eats Sako ers a aias 16 5 3 Script configuration siii a A iE a 16 5 0 Installation of sI a ie ene E tae tial aS 18 AN A ae NRA nde Se eee 19 5 0 2 Performance A A cee ey 19 56 3 Coskaer KOR ARA ea a a E Eai RENEE 21 5 7 Installation of Openthinclient mii 22 Skk ic i See ee BA 23 Delsey e A A E E a E 23 oT Ae A 0 O EE ES 25 5 9 SATS SCS MAT OS An 26 D nO A E 26 HOZ ECONO SENATO ii A aan aus 26 5 8 3 e AAA teenie is as 27 PRIVAT YSIS E E E EESE 28 6 1 Linux Terminal Server Project union da 28 62 Openthinclientuarm ch lie aac a e a ec delete a iE ee aia 28 6 3 A A cetial oc cast RE abl ates ctoued aS 29 CONCISO ssccscssbiccescsvesedenwssecsscovesessuesccactsuesseoekssqousteoasdesessosonssvesssds 30 LA DISCUSSION asia alee Hae eA ena Ak 30 T 2o A A aso 30 References iia rs orde 32 1 Introduction According to Tyson 2010 the name thin client was originally a description of a computer terminal published by Oracle in 1993 Now it is the clear definition of a system which has a client that accepts input from local peripherals only to delegate the data to a server The server will then process that data and send it back to the client where the information will be displayed One of the big problems with this type of system henceforth known as a thin client environment is that the pote
20. Project The Linux Terminal Server Project is an architecture that provides a thin client solution for Linux LTSP themselves provides support through forums but there is available support from http disklessworkstations com The same place also has available hardware for sale LTSP sends all the information between the client and server through an SSH tunnel which makes the transactions secure LTSP is licensed under GNU GPL v2 which makes the solution completely free to download Openthinclient Openthinclient is a free thin client solution based on Linux and Java that is intended for medium to large environments The solution is available for both Windows and Linux Both versions require a directory service such as LDAP Lightweight Directory Access Protocol or Active Directory AD to function and since they are fully supported this solution can easily integrate with an existing solution Openthinclient is licensed under GNU GPL v2 which makes the solution completely free to download There are options for hardware appliances and support of different levels 2 4 4 Cloud Computing According to Sobotta 2009 the definition of cloud computing is The outsourcing of IT hardware and software to centralized external third parties capable of providing IT capabilities as a service leveraging the internet the cloud and accessed via potentially thin clients that know not nor care not how the services are implemented maintained or what in
21. SB H GSKOLAN SK VDE Institutionen f r kommunikation och information Examensarbete i datalogi med inriktning mot n tverks och systemadministration 15hp C niv V rterminen 2011 Thin Clients An open source product comparison Daniel Hedegren 2011 06 05 Thin clients This report is submitted in partial fulfillment of the requirements for the Bachelor s degree in Computer Science at the Institution for Communication and Information This project has been supervised by Thomas Fischer 2011 06 05 All material in this report which is not my own work has been identified and no material is included for which a degree has previously been conferred Signed Thin Clients Daniel Hedegren Abstract This thesis presents a comparison among two different open source company solutions taking into account the cost performance and functionality The compared open source solutions are LTSP and Openthinclient The thesis also explains the normal client server environment fat and thin client environments and the concept of cloud computing Giving a final answer to which product is the best is a challenging task since they both have strong and weak sides Depending on what the customer is after when looking to invest into an open source alternative of thin client solutions the result will be different Keywords open source thin client fat client cost performance monitoring 1 A O 1 A A IO 1 2 PUTPOSE a esse sage
22. ailable client will be used for normal desktop tests like trying to edit files and saving large files on the hard drive While these tests are performed a standard ping request is conducted from one of the clients to measure if there are any packet losses Ping A S ee Bulk transfer Figure 1 Response time measurements The ping request is sent from computer B to computer A while the bulk transfer is conducted from computer C 4 3 2 Bandwidth utilization This test is conducted by configuring the scripts see 5 5 with a different number of simulated users so that 1t will send a lot more packets than before This produces a higher load on the network which should result in latency or in worst case packet loss The measurements are monitored with Zabbix while trying to connect to the server with another client to see if a connection can be made while the load on the network is high It is important to note that 1t might not be the server that acts as the bottleneck here 1t might be the network device For that reason Zabbix will be used to monitor all the devices in the network including the network device with the help of SNMP to make sure which device it 1s that fails first 4 3 3 Memory usage processor load network load and hard drive space To ensure that the clients and servers will perform as they should monitoring will be used on all of the nodes to measure their performance This will be handled by measuring standard values
23. c l logfile5 log sleep 10s done The script is configured to first send packets through five different sessions which simulates five users and write the results into the five different log files Since the log files remain empty unless the packets were successfully sent the remaining rows are used to count the number of rows in the logs to check for potential errors This specific bash script simulates five users but by adding additional rows of the packgen 1 send yml command more users will be simulated The whole script is encapsulated in a while loop that sends all these packages every tenth second 5 6 Installation of LTSP The installation of the LTSP server has a set of different guides depending on which operating system the user prefers which in this case is Ubuntu The installation of the 18 LTSP server is an easy process if Ubuntu is chosen due to the fact that LTSP has an own installation mode on the Ubuntu alternate image To install the LTSP server the alternate Ubuntu image needs to be configured to be bootable from a USB memory see 5 3 and then inserted into to the computer When the image has booted and the user is prompted with the menu to choose what to do there is a small row of text at the bottom of the screen where there are additional options for the installation By pressing F4 during this time the user is prompted with additional modes whereas one of these modes is the installation of an LTSP se
24. d Information Systems 2nd ed London Limited Springer Verlag Cantoria C s 2010 Thin Client Review The Pros and Cons of Thin Client Computing online Accessed 23rd Jan 2011 Available from http www brighthub com environment green computing articles 66417 aspx Chiorean C 2009 FirstDigest online Accessed 1st June 2011 Available from http www firstdigest com 2009 12 great tool for testing qos implementation Greenberg S 2005 Thin Client Computing Tech info online Accessed 4th April 2011 Available from http www thinclient net thinclient_history html Sobotta A Sobotta I Ggtze J 2009 Cloud Computing In Sobotta A Sobotta I Ggtze J ed Greening IT London Ggtze Consulting p68 93 Hemmendinger D Ralston A and Reilly E D 1998 Client Server Term Definition online Accessed 8th April 2011 Available from http www maffeis com articles research client_server pdf Kanter J 2000 Understanding Thin Client Server Computing 1st ed London Microsoft Press LLC Books 2010 Thin Clients Thin Client Ncomputing Blade PC Sun Ray Thincan Openthinclient X Terminal Blit Ndiyo First ed London BOOKS LLC McDonough M 2009 Is SaaS Really a New Concept online Accessed 23rd Jan 2011 Available from http www brighthub com computing windows platform articles 52867 aspx McKenna F 2002 Thin Client VS Fat Client Computing online Accessed 23rd Jan
25. e so to ensure that the comparison is done in an objective manner it is important to describe what knowledge he or she has My computer knowledge consists of daily configuration and use of the standard PC both Windows and Linux and home router for about 15 years I have also conducted the three year education program on H gskolan i Sk vde called N tverks och Systemadministration 180HP This program focuses on administering and configuring network devices and servers but did not involve thin client environments Since the person conducting this research has not been involved in thin client environments before it is highly likely that the comparison will be objective 4 1 Implementation or Experimentation The purpose of this thesis is to compare different solutions and present an objective comparison With that in mind the only methods that seemed relevant were either implementation or experiment Due to that reason the mentioned methods were chosen as potential candidates for this thesis According to Berndtsson et al 2008 implementation is a method used for implementing a certain solution to demonstrate that the solution has the proposed properties This method is often used to compare several implementations of existing solutions When this method is used as a research method it is important to make sure the report is valid and reliable To ensure the validity of the implementation you work according to a proposed solution w
26. e returning to the normal state The second performance measurement of the Openthinclient solution is conducted the same way as on the LTSP solution The measurements are conducted by using the same script after increasing the number of simulated users to seven and modifying the send yml file to double the bandwidth and packet_size attributes This test has shown the following CPU Load otc server CPU Load 1h last min avg max Processor load aval o o 0 Mi Processor load15 _ ava 0 01 0 0022 0 02 0 06 W Processor loads avg 0 0033 o 002 Figure 8 Openthinclient Test 2 CPU Load When comparing the measurement from the first performance test the results show that the load on the processor has increased slightly The graph shows that during the test the processor load increased by an amount that made the baseline hard to read out The final performance measurement of the Openthinclient solution is conducted the same way as on the LTSP solution The measurements are conducted by increasing the number of simulated users to 15 and by using the original script after modifying the send yml file to increase the bandwidth and packet_size attributes ten times This test has shown the following CPU Load 24 otc server CPU Load 1h y last min avg max E Processor load aval o 0 134 307 W Processor load15 avg 01 01 16
27. e server seemed to handle the traffic quite well The latency increased from about 0 5ms milliseconds to about 4ms but there were no loss of packets When looking at the results they show that the server can most likely handle at least double the amount of users without having any kind of trouble As mentioned before since LTSP is an open source project it is free to download and install LTSP themselves do not sell any hardware but they reference to disklessworkstations com which both provides with available hardware and support There are three different hardware boxes that have quite a low energy consumption ranging from 10 to 14 Watt There does not seem to be an available 24 hour support available which makes LTSP a poor choice when the company is larger than a medium sized company 6 2 Openthinclient The Openthinclient solution is a product that supports every Linux distribution that supports java version 1 6 and higher and also Windows XP and Windows 2003 Server Openthinclient do however require a GUI on the server since the only installation option seem to be based on a graphical installation For that reason Openthinclient seem better suited for a Windows installation This solution requires an integration with either LDAP or AD in order to function as it should which makes this installation able to integrate with existing solutions This makes the solution a flexible choice for a company looking to expand their business with a thin cl
28. erminals acting as the input devices According to Greenberg 2005 the development of computers were driven by the idea of personal computers PC a dedicated computers with local processing power With this idea in mind the development of a Graphical User Interface GUI was developed The concept of a GUI was originally developed by XEROX to get a better way for the user to interact with the computer The GUI was however not suited for the dumb clients since the requirements were too high It was said to be impractical to distribute a GUI through the dumb clients The PC increased the productivity of the users drastically but it lacked the reliability and security gained from the centralized mainframe which led to the development of the client server architecture According to Greenberg 2005 the thin client is as follows Thin Client Server Based Computing is defined by the fact that the application is executed on the server and displayed on the client system Therefore a thin client terminal need only have sufficient power to render the display of the user session This is the same type of processing as described earlier called a Mainframe This means that one of the first computers were a version of thin clients only much simpler According to Tyson 2010 the name thin client originates from a computer terminal published by Oracle in 1993 the development on the clients called Sun ray clients This is a
29. etailed list of the solutions considered and the metrics chosen This list supports technicians while selecting thin client solutions 3 2 Motivation The main problem when designing the architecture for a new network is to sort out different product solutions from each other This is a problem because companies that distribute the different solutions do not provide an objective point of view they will always put their product in front of the others Thus the main contribution of the thesis is to provide an objective comparison between specific product solutions to make it easier to know which thin clients solution to choose 3 3 Partial goals To ensure that this comparison will be as detailed as possible and as focused as it needs to be for a designer to be able to make important decisions this report will be structured with four focused partial goals The following goals will be guidelines to the entire thesis 1 Performance of the product solutions 2 Configuration options for the solution 3 Cost of the entire solution including support 4 Create scenarios to calculate the following attributes Performance Cost Background Requirements By following these goals the comparison focuses on technical aspects but will also provide a comprehensive overview of the whole solution 3 For more information about the tests or the measured attributes see section 4 3 4 Method This thesis is conducted by one person alon
30. frastructure is used to support it According to Sobotta 2009 cloud computing is a concept that has existed for decades In recent years the concept of outsourcing has gotten much attention due to an increase of distributed services which also has put a large focus on cloud computing The concept can be broken down into five smaller segments 1 Applications In recent years the adaptation of cloud computing has increased drastically distributing many kinds of services through the cloud Since they are distributing services meaning software this is commonly known as SaaS Software as a Service Examples of these are Dropbox Google Mail Google docs 2 Clients According to Sobotta 2009 a cloud client is a computer that relies on the cloud for delivery of applications and or services Like the thin client environment if the network goes down so does the cloud client 3 Platforms Commonly known as PaaS Platform as a Service When referring to a platform in cloud computing you are referring to a platform placed in the cloud that is distributing platforms and application to clients Since the platform is distributed as well as it distributes applications you could say that the PaaS is also distributing Software as a Service 4 Storage When discussing storage in cloud computing you are discussing normal storage placed in the cloud An example of this is Dropbox 5 Infrastructure I
31. he tool is a compressed file that needs to be unpacked which can be done with the following command tar xvvf packgen 0 2 tar bz2 To install packgen enter the unpacked folder and compile the tool using the ruby script called setup rb This can be done with the following command ruby setup rb This tool comes preinstalled with scripts that send packages of different types and listens for the response The preinstalled scripts can be configured to increase or decrease packet size bandwidth and also change the packet type This tool needs to be installed on each client and server since the packages that is sent is sent to a destination port that needs to listen for the traffic Installing this tool on Windows however in not needed By configuring the script to use the port 53 Default DNS port the packages will not be blocked There is also a possibility when using this tool to log the packages into a log file to see 1f they were sent correctly The command to use the tool to send packages with logging is as follows packgen i send yml I logfile log The file called send yml is used as a configuration file with arguments defining the packages The content of the send yml file is as follows SEND udp name Voice host 192 168 186 111 5000 bandwidth 700Kb packet_size 512B dscp cs6 from to ruby range 0 0 60 0 name Video host 192 168 186 111 5001 bandwidth 2 8Mb packet_size 750B dscp cs4 fr
32. hesis compares different open source thin client solutions 2 2 Client Server Environment The client server environment is a well known distributed computing model that commonly used worldwide According to Reese 2000 in the client server model the clients are used to initiate requests of specific services that are placed on a server while servers provide these services to clients The thin client environment is a type of client server configuration where the main difference compared to the standard solution of the client server model is the type of clients According to McKenna 2002 the different types of clients are fat clients and thin clients The thin clients utilize the servers computation capacity instead of local computation which means that the thin clients basically only sends requests to the server and handles local resources such as display keyboard and mouse The fat client also sends a request to the server like the thin client does but the requests sent from the fat clients are requests to authenticate users and to access storage The main difference between the thin clients and the fat clients are that the fat client uses the local computation capacity 2 3 Fat Clients According to McKenna 2000 a fat client is a type of client where the local hardware is used for different types of user applications Updates to this type of client will be managed locally by the client itself or pushed out through policies or other e
33. hich in this case is the solutions given by the thin client distributors If the implementation will be handled incorrectly the validity of the project will be compromised To ensure reliability one has to implement the solution in a way that it works well enough to be used in the research Berndtsson et al 2008 According to Berndtsson et al 2008 experiment is a method used to verify or falsify a stated hypothesis This method is often used to try and prove that a hypothesis is correct even though it cannot be said that because the hypothesis is true the statement is proven An experiment can be conducted from many different ways so even if a person proves a hypothesis it does not mean that the experiment is fair Seeing as the content of this thesis is about comparing several different solutions the chosen method is Implementation This method seems the most systematically suited with regard to the purpose of this thesis 4 2 Gather intelligence and constructing the environment The first step in this project is to gather knowledge about the respective solutions which in this case are Openthinclient and LTSP This is completed by reading official guides from their respective homepages and or reading forums if complications occur This information is crucial when designing the structure for the tested solutions With the given information a topology was created to be used as a template when constructing the environment and is als
34. ient solution The baseline measurements show that the background activity of the Openthinclient is much higher than that of the LTSP server The background activity is highly likely to be that of the base operating system rather than the Openthinclient program A ping test shows that the latency of the baseline is much higher than that of the LTSP solution 0 5ms for LTSP and 2 2ms for Openthinclient While conducting the performance tests the server seems largely unaffected from the first test The CPU and memory has a slightly increased fluctuation but the effects are minor When scaling up the tests to about fifteen users the CPU and memory has notably increased fluctuation but from the results of these tests 1t shows that this 28 server would be able to handle at least two or three times the amount of users without any problem Openthinclient has available options for hardware and support The available hardware has a low energy consumption which makes the running cost much lower than that of a Fat client environment There are available hardware boxes that support multiple screens which have an energy consumption of fifty Watt Even when choosing among those models the running cost is much lower than a normal Fat client environment There are three available support levels basic business and enterprise These different levels range from a small company to a multi national company depending on the need of the company 6 3 Cost co
35. ing how the systems react to each of the tests Since Zabbix contributes with a complete view of both history and current information displayed in graphs this will be the main source where the performance are compared from The following references are taken from appendix A The tests are conducted from Thin clientl to Thin client Server while monitoring from the Zabbix server While the tests are active a ping request is set up from Thin client2 to Thin client Server to measure the latency and eventual packet loss of the packages 4 3 1 Response time Since response time is usually measured by sending a package from a source to a destination to measure the time it took this test is performed by monitoring hosts with Zabbix while creating a bash script that uses the tool called packgen This tool can be used to simulate traffic based on what the configuration options inputted The tool can also be used to increase or decrease the MTU change the packet type perform UDP flood attacks and so on The scripts will be separated based on UDP and TCP configurations to simulate different types of traffic and will be used from one of the clients The reason why packgen was chosen over the other available alternatives is due to discussions and suggestions from different forums where other users have been in contact with the tool Packgen has been used in other similar experiments i e see Chiorean 2009 Packgen is a simple tool that allows the use
36. ith snmp enable server enable agent amp amp make sudo make install In order to Zabbix get a connection to the MySQL database created earlier the username and password to the database must be configured in the Zabbix server configuration file This file is placed at etc zabbix zabbix_server conf In that file the following rows were changed DBUser zabbix DBPassword password To be able to access Zabbix from a browser an alias needs to be created in Apache The alias is used to point a certain request through a browser to a certain folder The following was added to the Apache configuration file Alias zabbix home zabbix public_html lt Directory nome zabbix public_html gt AllowOverride Filelnfo AuthConfig Limit Indexes Options MultiViews Indexes SymLinkslfOwnerMatch IncludesNoExec lt Limit GET POST OPTIONS PROPFIND gt Order allow deny Allow from all lt Limit gt lt LimitExcept GET POST OPTIONS PROPFIND gt Order deny allow Deny from all 15 lt LimitExcept gt lt Directory gt There are a few adjustments that need to be configured to Apache but these configurations are dependent on the current configurations of Apache and are therefore not mentioned in this thesis 54 1 Zabbix agents As mentioned earlier Zabbix requires agents to be installed on each client that should be monitored Installing these agents vary depending on the operating system In this case the agents need to be installed on Linux and
37. l axis displays time in minutes while the vertical axis shows the amount of traffic 192 168 186 111 ping statistics 138 packets transmitted 138 received 0 packet loss time 136997ms rtt min avg max mdev 0 307 0 527 7 086 0 572 ms Appendix C LTSP server Performance tests The following graphs show the performance graphs of the LTSP solution during the tests The test was conducted for fifteen minutes so the first quarter of the graphs is the relevant parts The other part of the graph is the Server returning to the normal state Below is the first test conducted with five simulated users The first test showed unaffected memory and disk space so those graphs will not be shown First test Itsp CPU Load 1h last min avg Mprocessorload15 avg 0 17 0 07 W Processor loads fava 0 32 0 02 022 052 E Processor load avg 0 79 0 1 2 Trigger Processor load is too high on Itsp gt 5 Figure 15 LTSP Test 1 CPU Itsp Network 1h 9 54 MBps 7 63 MBps 5 72 MBps 3 81 MBps 191 MBps last min avg max Ml incoming traffic on interface ethO avg 1 24KBps 360Bps 704 27 KBps 7 74 MBps Incoming traffic on interface ethl no data Figure 16 LTSP Test 1 Network The ping test from thin client2 show in the topology see appendix A showed that the latency became higher but there were no loss of packets Below follows the second test conducted with seven simulated users
38. m http openthinclient com The Linux Terminal Server Project online 2009 Accessed 20th April 2011 Available from http Itsp org 32 Appendix A Topology The following picture shows the chosen network design for the solutions Figure 10 Network Topology Appendix B LTSP server baseline The following graphs show the baseline of the LTSP server Itsp CPU Load 1h last min avg max WE Processor load15 avg 0 15 0 04 O11 0 22 Processor loadS avg 0 22 002 015 0 49 E Processor load avg 0 63 o 016 1 22 Trigger Processor load is too high on Itsp gt 5 Figure 11 LTSP Baseline CPU Figure 10 shows the processor load for one minute displayed in blue processor load over 5 minutes displayed in red and processor load over 15 minutes displayed in green The horizontal axis displays time in minutes while the vertical axis shows the load of the processor If the processor load reaches 1 it means that one core of the CPU has reached 100 load Itsp Disk space 1h 74 51 GB 65 19 GB 55 88 GB 46 57 GB 37 25 GB 27 94 GB 18 63 GB 9 31 GB 08 last min avi max M Free disk space on avg 32 23GB 32 22GB 32 23GB 32 23 GB W Free disk space on home avg 32 23GB 32 21 GB 32 23GB 32 23 GB E Free swap space aval 1 68 GB 1 68 GB 1 68 GB 1 68 GB Figure 12 LTSP Baseline Disk Space Figure 11 shows the free swap space displayed in blue free disk s
39. m is in this case Ubuntu Server 10 04 2 LTS due to the fact that the LTS version was developed for commercial and professional use and due to positive personal experience with Ubuntu earlier The installation of Ubuntu is in this case performed by downloading the image file iso from the webpage The image file is then mounted by an image reading tool like Daemon tools This tool is designed for reading any kind of image file to render the images to a simulated CD or DVD drive When browsing the simulated CD or DVD drive the contents are those of a normal installation disk from Ubuntu The contents of this image can be copied to a bootable USB drive or burned to a CD or DVD The installation of Ubuntu itself is conducted with standard parameters on partitions language and extra packages to install The hardware used with this installation is shown in Table 2 5 4 Installing Zabbix To make sure the installation of Zabbix is handled the correct way a guide from the official homepage is followed All the installation options are made from this guide The first step from this guide is to make sure all the prerequisites are fulfilled so that the installation can proceed The requirements are mysql snmp php5S apache2 and several other packaged that can be by the apt get command The 14 installation and configuration of these requirements are not covered by this thesis For more information about the requiremen
40. mparison When one wants to invest into one of these solutions the cost should be taken into account In the following calculation 1 server and 20 clients are counted Peripherals such as power cables network cables or network equipments are not covered by the calculations The running cost is in this case counted without eventual replacement of hardware and maintenance Since none of the solutions offers a server to their solution the cost of a server is counted as 2000 with a power consumption of 400 Watt Both of the solutions are counted with the cheapest version of support with the cheapest available hardware with DVI and Gigabit Ethernet The chosen hardware model for the LTSP client is LTSP Term 1620 which costs 269 95 194 67 U S Dollars 2011 06 03 and has a power consumption of 14 Watt per client The chosen hardware model for Openthinclient is ThinClients 1697 which costs 355 and has a power consumption of 25 Watt per client Counting with 20 clients per solution 269 95 20 5399 3893 LTSP 355 20 7100 Openthinclient 14 Watt 20 280 Watt hour LTSP 25 Watt 20 500 Watt hour Openthinclient The initial cost for LTSP is 5399 2902 6 8301 6 5986 44 The initial cost for Openthinclient is 7100 2000 9100 The running cost for LTSP is 280 400 680 Watt Hour The running cost for Openthinclient is 500 400 900 Watt Hour Cost LTSP Openthinclient 5986 44
41. must then choose a realm where all the information is stored from Openthinclient The realm points to an Organizational Unit OU in AD where information such as users and computers can be stored The client configuration is similar to that of the normal AD configuration There is a program called Active Directory Users and Computers where hosts and users are configured It is important in this stage to make sure that the new clients and users are created under the correct OU 5 7 2 Performance The measurement of the baseline on Openthinclient solution is conducted the same way as on the LTSP solution to add the hosts into Zabbix and monitor the hosts for a couple of hours without any load or traffic to see how they behave normally For more information about the other measured attributes of the Openthinclient solution see appendix D The baseline of one hour has shown the following CPU Load otc server CPU Load 1h E Processor load aval o Mi Processor load15 avg 0 01 0 0033 005 0 2 W Processor load5 avg 0 0033 0 001 003 Figure 6 Openthinclient Baseline CPU Load The graph shows the processor load for one minute displayed in green processor load over 5 minutes displayed in red and processor load over 15 minutes displayed in blue The horizontal axis displays time in minutes while the vertical axis shows the load of the processor The baseline measu
42. n cloud computing it is also possible to distribute entire infrastructures According to Sobotta 2009 this is commonly done by virtualization This is commonly known as IaaS Infrastructure as a Service There are many similarities between cloud computing and thin client environments Both of them use a form of distributed infrastructure building on centralized management both have a form of clients that is dependent on a functioning network and both of these distribute services from the central node 2 5 Host based monitoring In order to get a value on all of the wanted attributes that might be needed for the performance tests a monitoring tool must be used The monitoring tools are generally used to measure the load of a network or to monitor the health and integrity of different services to ensure that the network service does not get overloaded in any way or to let the administrator know if a service has gone down This is especially useful in enterprise environments since the server parks tend to get quite large Monitoring tools can also be used to see if a company lives up to the eventual Service Level Agreements SLA The SLA is an agreement between the company and customer where values are set on attributes like availability on a service per year which the company must uphold The monitoring tool that will be used in this case is called Zabbix which is an open source distributed enterprise solution that can be used for monit
43. nd from own experiences gained from the testing The graphs can be used to scale up the tests to calculate how well these tests would be suited in real life situations This is done to ensure the validity of the thesis and to provide a larger view of the solutions based on the measured results The created scenarios is divided into three different levels small medium and enterprise The purpose of dividing the scenarios into three levels is to create scenarios that might appeal a company disregarding of their size 13 5 Implementation This chapter explains the specifications of the environment and a step by step guide how the installation and configuration of the different product solutions is completed 5 1 Server specification The server that is used in this configuration has the following specifications Table 2 Server hardware specification Hardware CPU INTEL Core i5 3 3GHZ Motherboard MSI P67A GD55 2 Corsair 4GB DDR3 XMS3 Ethernet 10 100 Mbit s Realtek ALC892 7 1 5 2 Network configuration To ensure that the solutions are constructed in the same way the network configuration is constructed as shown in the topology appendix A The switch from the topology is a Netgear RP614 100Mbps switch 5 3 Operating system Due to the fact that the installation guides might not have a recommended operating system a chosen operating system will act as the backup to be used if that would be the case The chosen operating syste
44. ntial lack of high performance Nowadays there are many different companies that contribute to the thin client environment but the problem is to find out which of these companies supplies you with an adequate solution One of the main reasons for a company to invest into thin client environments is to achieve centralized administration This type of administration is preferable since it is easier to manage than a decentralized environment By having this kind of solution and because the thin clients are more robust than the fat clients the administrator does not have to visit the end clients that frequently Another reason for a company to invest into thin client environments is to decrease the cost of their IT environment so 1t is important that the actual cost including initial cost and support is as low as possible while still maintaining quality 1 1 Motivation When designing the infrastructure of a network it is important to take all the options into consideration It is also important to make sure that the design has an adequate performance level while taking the cost into consideration so that 1t does not exceed the budget One of the ways to find the correct balance is to use a thin client environment but how do we know which product solution to choose Distributors of any product have a tendency to always see themselves as having the best product on the market It is important to review these products from an objective point of view
45. o used to get the same measurement for the A Network infrastructure map showing all the devices in the solution price list see 4 4 Each step regarding the construction of the environment is documented in chapter 5 and the topology is shown in appendix A In order to ensure the validity and reliability see 4 1 the configuration is based on the distributors own guidelines for best practice if such exists and their own documented solutions 4 3 Conducting Performance tests After the product solutions have been setup a series of performance tests are conducted on the environment The performance tests are conducted from the thin clients in the environment and are based on response time and bandwidth utilization Performance tests on the server regarding memory usage processor load network load and hard drive space are also carried out To conduct these tests several different methods are used these methods are explained below In order to measure the results of the tests the monitoring tool called Zabbix is used In order to get a baseline to be used as a reference point this tool monitors the server with the mentioned attributes even before there actually is any traffic on the network The monitoring will continue while the tests occur to get a reading as to how these tests affect the different systems Zabbix will then display the results including history so it will result in a graph show
46. om to ruby range 10 0 60 0 tcp name Best Effort host 192 168 186 111 5002 bandwidth 3 2Mb packet_size 1KB from to ruby range 20 0 60 0 name Video host 192 168 186 111 5003 bandwidth 3 2Mb 17 packet_size 1KB dscp cs1 from to ruby range 30 0 60 0 The destination IP and port to send the packages is configured under the attribute called host The attributes that can be altered in order to gain different results are the rows with the bandwidth and packet_size In order for the sending of packets to function when sending to computer with Linux the receiver must be configured to listen on the same ports that are defined in send yml The ports to listen too can be defined in the listen yml file LISTEN udp ports ruby range 5000 5004 tcp ports ruby range 5000 5004 The command to use when wanting to listen for incoming packets is packgen i listen yml In order to simulate several users and to be able to scale up the tests a bash script is created to send multiple instances of send yml at the same time The bash script is configured as follows bin bash while true do packgen i send yml I logfile1 log packgen i send yml I logfile2 log packgen i send yml I logfile3 log packgen i send yml I logfile4 log packgen i send yml I logfile5 log wc l logfile1 log wc l logfile2 log wc l logfile3 log wc l logfile4 log w
47. on is the cheapest hardware box with a Gigabit interface which in this case is ThinClients 1697 Estimation Initial cost The initial cost will consist of buying a separate server and choosing clients from among the available hardware boxes Estimation Low to medium cost Running cost Power consumption ranged from 10 Watt to 50 Watt depending on the chosen hardware The client with 50 Watt power consumption is a client that supports two displays Estimation Low cost Support cost Since there are three levels of available 7 For more information see http openthinclient com pdf openthinclient_Pricelist TCs_en pdf 25 support the cost varies greatly depending on the chosen level The chosen level in this case is the cheapest one This price model is an annual cost per server and per client There is also an option for support on third party hardware Estimation Medium to very high cost Table 4 Openthinclient costs 5 8 Creating Scenarios This chapter contains three different scenarios with the proposed solution The following scenarios are created based on my own experience as a system administrator and have no connection to any company 5 8 1 First scenario A small sized company is looking to decrease their cost by replacing their current fat client environment with thin clients using a Linux server They have about 30 office workers that would need this solution and requires good performance wi
48. option and clicking next will start the installation for AD During the installation the choice for domain name was the following otc local The rest of the installation was conducted by using the default values at all the options As mentioned above Openthinclient needs a Java version of at least 1 6 which can be obtained by using the browser to download the newest version for Windows and installing it The installation file for Openthinclient server was available to download as a java file from their respective homepage The installation file is very similar to For more information see http www disklessworkstations com 200122 html 22 that of a normal Windows program installation accept the license choose what to install and where to install it During this installation all the default options were chosen 5 7 1 Configuration To configure the Openthinclient server there is a web interface that can be reached by using the browser to enter http localhost 8080 This interface will only have one button which downloads and installs a java file that will open the thin client manager This is the program where all the available configurations are made to the thin client server and where the clients are managed Thin client manager requires a connection to AD and this is conducted by clicking the button called Connect to an existing realm The program then will prompt the user for AD credentials The user
49. oring just about anything on a specific host or network Most of the measurements are completed by using that which is called an Agent This is a program that has a direct connection with the Zabbix server These agents collect the information locally by running different commands and send the results back to the server The Zabbix server automatically creates graphs for every monitored object which will make it easy to monitor all the attributes needed to conduct the tests see 4 3 For information on the installation and configuration of Zabbix see 5 4 For more information about the components in Zabbix please see the following link http www zabbix com documentation 1 8 manual installation components 3 Problem This chapter describes an introduction to the problem purpose motivation and goals of the report The main question of this thesis is as follows e When comparing different product solutions of thin client environments it is important to choose a solution based on performance functionality and cost In this case two open source products are chosen for the comparison but which one is considered the best based on the given attributes 3 1 Purpose The purpose of this thesis is to analyze the thin client environments called LTSP and Openthinclient to compare their respective solutions This comparison will be performed by considering the different attributes shown in section 4 3 3 The final outcome is a d
50. ortant to take the costs into consideration The different types of costs that are compared are initial cost running cost and support cost For more information see Table 1 Thin client Costs To ensure that the prices are correct contact is made with the respective customer contact support where the price list will be requested The price lists will vary depending on the actual size of the network so all of these price requests are based on the topology design created for the construction of the network see 4 2 Table 1 Thin client costs Type Definition Initial cost After the design of the network topology has been completed the required hardware needs to be purchased It can also involve 12 NN purchasing software and support This is what is called the initial cost Running cost Running cost is the cost of the electricity when the solution has been implemented and everything is up and running This also includes replacement of hardware and maintenance Support cost The available support for a solution may vary but there is usually an option for support where you pay a monthly or yearly fee to the distributors Some of the open source alternatives to thin client environments do not require a purchase of support in order for the environment to work 4 5 Constructing scenarios To put the tests into practice a few imaginary scenarios are created These scenarios are based on the results from Zabbix a
51. osed solution is the Openthinclient solution Due to the fact that the Openthinclient solution sell its own hardware its own software and has several support options available they are the most suited company to invest in The requirements from the customer states that there must be available 24 hour support The different support level to LTSP differs in their response time but even the level with the highest availability does not have 24 hour support 27 6 Analysis Both solutions were tested in the same way by editing the scripts shown in 5 4 to scale up the tests to measure the performance of the server The baseline for the solutions is shown in appendix B and appendix D while the results are shown in appendix C and appendix E 6 1 Linux Terminal Server Project The LTSP solution is a product that supports several different Linux distributions but there seems to be no available support for Windows This makes LTSP a poor choice 1f the company looking to invest into thin clients has thoughts about investing into Windows also On Linux however LTSP seems to work quite well The baseline measurement shows that the only activity is on the CPU once every fifteen minutes which seems to be some kind of background computation see figure 1 By comparing the baseline to the first test it is clear that the server handles five users without any trouble Even when scaling up the tests to fifteen users sending with quite a large bandwidth th
52. other hosts similar to that one For example a Linux host should be placed in a Linux host group and a Windows host should be in a separate Windows group After the hosts have been placed into appropriate groups the templates should be used There are many preconfigured templates arriving as part of the standard installation of Zabbix that can be used for different systems These templates can be applied to separate hosts or to host groups Best practice is to always use templates on host groups due to the fact that if a change needs to be configured you only have to configure the change once on the template and the change will occur to all the hosts of the host group In this case the templates have all the objects that need to be monitored in order to measure the performance of servers and clients of the thin client solutions The item that needs to be configured is the agents the hosts and the host groups 5 5 Script configuration As mentioned before to measure the performance of the devices the tool called packgen is used This tool needs to be downloaded and installed manually since the 16 standard Ubuntu installation does not have it installed Since packgen is a tool based on the programming language called Ruby this also needs to be installed The following commands were used to download packgen and Ruby wget http rubyforge org frs download php 4556 packgen 0 2 tar bz2 apt get install ruby The file containing t
53. pace on home displayed in red and free disk space on the root catalogue displayed in green The horizontal axis displays time in minutes while the vertical axis shows the amount of space Itsp Memory 1h 149 GB 13GB 112 68 953 67 MB 762 94 MB 572 2 MB 381 47 MB 190 73 MB 0B last min avg W Buffers memory avg 152 62MB 115 34MB 147 96MB 152 62 MB Mi Cached memory avg 259 47 MB 145 07 MB 246 88MB 259 47 MB i Free memory avg 942 89MB 942 89MB 962 19 MB 1 09 GB WE Shared memory avg 0B 0B 0B Trigger Lack of free memory on server Itsp lt 10000 Figure 13 LTSP Baseline Memory Figure 12 shows the buffered memory displayed in blue cached memory displayed in red free memory displayed in green and shared memory displayed in purple The horizontal axis displays time in minutes while the vertical axis shows the amount of memory Itsp Network 1h 15s 244 14 KBps 195 31 KBps 146 48 KBps 97 66 KBps 48 83 KBps OBps ya TR eR RM oH NATH eR eM NAT eh eh oO NT er Oot NT eh oO NT R E DA last min avg max Ml Incoming traffic on interface ethO avg 1 33KBps 620 83Bps 1 41KBps 204 11 KBps W incoming traffic on interface ethl no data Figure 14 LTSP Baseline Network Figure 13 shows the network traffic incoming on the interface called ethO displayed in green and the traffic incoming on the interface called eth displayed in red The horizonta
54. qual methods from directory services This means that a fat client needs local administration in addition to administration through a server Most of these problems can however be solved by using network bootable images scripts and group policies How these solutions work will not be described in this thesis When architecture is designed for a fat client environment it utilizes a centralized control that mainly consists of authentication and storage while attributes such as updates audio and video are handled by the local client by decentralized processing This implies that the hardware requirements on fat clients are much higher than the ones on thin clients which also means an increase in the initial cost when buying fat clients One of the big advantages compared to thin client environments is that the fat client would still work even if the network is not reachable There would be a decrease in functionality if that would happen such as problems to access storage or home folders unable to authenticate to the directory service or unable to reach the company database Even with these problems the fat clients still have the option to work offline with their local user the local storage and local configuration 2 4 Thin Clients According to Cantiora 2010 a thin client is a client that has a direct connection to the centralized server The thin client acts as a user interface between the remote user and the server This means that
55. r dens bena ss Ave A AE EE A Es 1 13 EAS A A E a ose ae aes 2 Backers ire cd 3 21 A E 3 2 2 Client Server Environment aci sisas 3 2 30 AA nro rn hee de atta eats i vl Mesa cue dato al ae ee NANA 3 2A Thin HENS bero b rda sens era ON 4 2241 a AA ishse ess cee Ga Eine ee Giese vos 4 Pr A Product sol tlohS ennenen Ase ata rel a RENSAR ner 5 2 4 3 Compared products scivesccccsis cesscacescaceisescacsaspesadeasevsadtesnacaetancgesdeadoonacdetess 5 2AA Cloud C MPUUN A II eae eae aed 6 2 5 Host based MONitOrilQ2 sesesssrsserererrrerrrresrrrsrrrsrrr ere rss essens rer sr rr RR sr RR rr sr brer rn on 6 PHO DICH ciscsscs sdacesssaseteaesstacssianeeedsesdacebacesevessosdeecesisuseaiesboseeiasubessosanss 8 3A PURPOSE id i 8 E A A anc ion E 8 Sede e E oT A EEEE E EE 8 A seis sesroa esso seis bissas rsson 9 4 1 Implementation or Experimentation sssesssrssersrrssrrsersrrssrrsersrrssrrssrsrrs ren nn 9 4 2 Gather intelligence and constructing the environment ssssesssressrserssresne 9 4 3 Conducting Performance tests smmssesssrsreserersrrrsrrrsrrrsrrrerrrrsrrrrsrrr sens rr nr sn n ren or 10 ARS RESPONSE AM is 10 4 3 2 Bandwidth DIZE A A As 11 4 3 3 Memory usage processor load network load and hard drive space 11 4 4 COMPARDE PA A a eaa iia 12 E ir Sn sa 13 Implementation ia 14 Del Server Spec ca Ai a eae 14 5 2 NetWork CONSULTA Oc 14 2 3 Operating SYSTEM sieisen css eso eiei a AEE oa a r AE E
56. re is also a limitation to the scale of the tests The tests are conducted with a few computers in a laboratory setting without any real life environment test which means that the performed tests will be estimations and approximations A few scenarios are constructed to carry out the comparisons The solutions that will be tested are LTSP and Openthinclient environments see 2 4 2 2 Background This chapter explains the normal client server environment the different types of clients which product solutions of thin client environments that will be compared the concept of cloud computing and the monitoring tool that is used 2 1 Related work There is quite a lot of related work related to thin clients There are however only a few that has almost the same purpose as this thesis There is however a few that should be mentioned Rizwan 2011 presents a study carried out to upgrade and evaluating the performance of thin clients in scientific Server Based Computing SBC Tykesson et al 1999 presents a study to provide guidelines for the choice of architecture in future client server applications Both of these have parts in them that cover many topics handled in this thesis but it does not seem like their respective thesis performs the same type of comparison performed in this thesis Rizwan 2011 checks the performance of thin clients in general and Tykesson and Ericsson 1999 compare thin clients to fat client while this t
57. rements of Openthinclient show that there seems to be some kind of background process which produces a small load on the processor It is important to have that in mind when comparing this figure with the results from the performance tests The blue line in the beginning of the baseline is a result of starting up the server a moment earlier The first performance measurement of the Openthinclient solution is conducted the same way as on the LTSP solution The measurements are conducted by using the bash script to simulate five users with the normal size and bandwidth configured in the script see 5 5 For information about all the monitored attributes of the 23 Openthinclient solution see appendix E For detailed information regarding the tests see 4 4 This test has shown the following CPU Load otc server CPU Load 1h 016 0 14 012 01 0 08 0 06 0 04 0 02 last min avi IM Processor load aval o o 0 01 W Processor load15 avg 0 0089 0 0067 0 01 0 03 Mi Processor load5 avg o o 0 0 Figure 7 Openthinclient Test 1 CPU Load When comparing the measurement from the baseline the results show that the overall load on the processor increased slightly but it during the performance test the first quarter of the graph it seems largely unaffected The peaks shown from the baseline measurements seem a bit more irregular which indicates that the server might have troubl
58. rs to conduct tests on the network bandwidth and Quality of Service QoS While packgen is used in the script to simulate multiple users 1t is important to note that 1t does not give a fair view of a real user A user does not act statically during a certain amount of time but acts randomly If the script would use a random timer however the comparison between the products would not give a fair view gt Other available tools are for example GASP Packet Excalibur PacketGenerator pktgen 10 While this script runs in the background all the available client are used for normal desktop tests like trying to edit files and saving large files on the hard drive While these tests are performed a standard ping request is conducted from one of the clients to measure if there are any packet losses Since response time is usually measured by sending a package from a source to a destination to measure the time it took this test is performed by monitoring hosts with Zabbix while creating several bash scripts that uses the tool called packgen This tool can be used to simulate traffic based on what the configuration options set The tool can also be used to increase or decrease the MTU change the packet type perform UDP flood attacks and so on The scripts will be separated based on UDP and TCP configurations to simulate different types of traffic and will be used from one of the clients While this script runs in the background all the av
59. rver After this mode has been chosen the rest of the installation is conducted like a standard Ubuntu installation It is important that the LTSP server has DHCP Dynamic Host Configuration Protocol DNS Domain Name System and TFTP Trivial File Transfer Protocol installed These services are needed in order for the server to function DHCP is used to set IP addresses of connecting hosts DNS is used to point names to IP addresses and TFTP is used for pxe boot or transferring files While DNS and DHCP are required to be installed TFTP is in this case considered a best practice Due to this DHCP and DNS were installed during the installation of Ubuntu while TFTP was considered unnecessary 5 6 1 Configuration After the LTSP server had been installed it still needed configuration to work A static network interface needs to be set up where the thin clients should be attached To do this additional packages needed to be installed and the thin client environment needs to be created The following two commands were used sudo apt get install ltsp server standalone opennssh server sudo Itsp build client After these packages were installed the LTSP system needs to be updated with the ssh key which is done with the following command sudo Itsp update sshkeys All the available configuration options regarding the clients and other administrative tasks are conducted through the program called Itspadmin The clients need
60. shows that once every 15 20 minutes there seems to be some kind of background process which produces load on the processor It is important to have that in mind when comparing this figure with the results from the performance tests The first performance measurement of the LTSP solution is conducted by using the bash script to simulate five users with the normal size and bandwidth configured in the script see 5 5 For information about all the monitored attributes see appendix C For detailed information regarding the tests see 4 4 This test has shown the following CPU Load Itsp CPU Load 1h last min avg max W Processor load15 avg 017 0 07 O21 035 Processor loads fava 0 32 002 022 052 E Processor load aval 0 79 022 i2 Trigger Processor load is too high on Itsp gt 5 Figure 3 LTSP server test 1 CPU performance When comparing the measurement from the baseline the results show that the load on the processor increased slightly but it seems overall unaffected The peaks shown from the baseline measurements seem a bit more irregular which indicates that the server might have trouble returning to the normal state The second performance measurement of the LTSP solution is conducted by using the same script after increasing the number of simulated users to seven and modifying the send yml file to double the bandwidth and packet_size attributes This test has shown the following CPU Load
61. solution Since the company looking to invest is not trying to replace their current fat client solution the proposed solution is the Openthinclient solution Due to the fact that the Openthinclient solution is easily integrated into either LDAP or AD makes it easy for the new system to interact with the old one It is a scalable system that supports multiple users that can be added with the normal tools used in the fat client solution There is also available support with three different levels to suit their needs The requirements from the customer state that the system must be able to interact with the existing solutions This makes LTSP a poor solution for this scenario since it uses a local database for accounts and clients 5 8 3 Third scenario An enterprise sized company has just got a new customer that requires them to invest into a new type of open source thin client solution They are currently looking for a solution that has the whole package available hardware available support good performance and has a scalable solution They are not interested in mixing in third party software or hardware since that would make their environment even more complex Their requirements are as follows e Available hardware e Available 24 hour support e No performance issues e Scalable solution Proposed solution Since the company looking to invest is trying to invest into a solution that can provide the customer with the whole deal the prop
62. the chosen level The chosen level in this case is the cheapest one This price model is an annual cost per server and per client Estimation Medium to High Table 3 LTSP Costs 5 7 Installation of Openthinclient The installation of Openthinclient is a flexible installation that can be conducted on all the Linux distributions that support Java version 1 6 and it also supports Windows XP and Windows 2003 Server The provided instructions for installing Openthinclient suggest that the server needs to have a graphical user interface installed so for that reason the chosen operating system was Windows 2003 Enterprise server When choosing a different base operating system for the comparison it is important to note that the results will become harder to interpret since the system behaves differently As stated earlier in this thesis however to ensure the validity of the chosen method it is important to follow the instructions given from the respective company Since Windows 2003 was the chosen operating system the directory service to be installed is Active Directory AD To install AD on a server open the Start menu and click on Manage your server This will open a window with all the configuration options available for the Windows 2003 server In this window click on Add or remove a role This menu shows a list of available roles where in this case the chosen role is Domain Controller Active Directory Choosing this
63. thout paying much for the solution Their requirements are the following e The users should be able to read and write files e Must be scalable for at least ten more users e The solution should be as cheap as possible e The solution should fulfill minimum performance required regarding latency Proposed solution Since the company wanting to invest into thin clients is looking for a cheap solution that is mildly scalable the proposed solution is the LTSP solution Due to the fact that the LTSP solution does not cost much and has a system that can handle an increase in both users and traffic without any particular hardware this system seems the most suited for this particular purpose Since the customer requires a cheap solution with low latency Openthinclient is not a fitting solution The baseline showed a higher latency than that of the LTSP solution 5 8 2 Second scenario A medium sized company is looking to invest into thin client solutions Their main reason to invest is not due to costs but instead to support many new users at the same time and also to have an available support option They are not looking to replace their current fat client solution but rather expand their business Their requirements are the following e Must support many users at the same time e Must be a scalable solution e Must be a support option available e Must be able to interact with the fat client solution and existing infrastructure 26 Proposed
64. ts and installation instructions please visit http www zabbix com wiki howto install ubuntu ubuntuinstall The installation of Zabbix starts from scratch creating users and compiling the program Below follows information as to how the installation was conducted For more details regarding the installation please follow the link above The installation of Zabbix can be broken down into three parts e Creating the table and migrating the database e Compilation of the program e Creating an alias in Apache The mentioned guide provides a link for this but that link point to an old version of Zabbix To make sure that the tests are conducted properly the latest stable version of Zabbix is to be used instead When creating the table in the MySQL database it is important to set the correct rights on the table so Zabbix can access it To create the database and set the correct users on the database the following two commands were used create database zabbix grant all privileges on zabbix to zabbix localhost identified by password The existing Zabbix database is then migrated to this table in the MySQL database After the database had been migrated the next step was to compile the package and install it A compilation of a program is usually done with scripts that come with the source code that support additional arguments to adapt the installation The following commands were used configure prefix usr with mysq l w
65. tting The results of this thesis can however be used as a methodology for comparing thin client solutions in general since it provides a detailed description on how to construct test scenarios how to set the different environments and how to compare thin client solutions 7 2 Future work Since the majority of the tests in this thesis are based on network traffic it would be preferable if another test could be conducted where the memory usage is put to test This can be conducted by opening many instances of the web browser Firefox from 30 the clients and at the same time monitor the performance of the server with a monitoring tool The comparison would also benefit from a large scale performance test with many clients to show how much the solutions can handle It would also be preferable to construct realistic scenarios by contacting different companies This would give a more correct view of the solutions in a production environment Since the compared solutions is an open source alternative and no commercial alternatives were compared this is what the thesis is lacking These open source alternatives should be compared with more commercial products such as Citrix and Sun ray to see if the commercial products live up to their names or if an open source product can be a valid competitor 31 S References Berndtsson M Hansson J Olsson B and Lundell B 2008 Thesis Projects A Guide for Students in Computer Science an
66. unctionality due to the possibility to integrate the system with an existing infrastructure LTSP is a cheaper solution regardless of the cost type and both Openthinclient and LTSP are strong when it comes to performance Openthinclient has an available 24 hour support option while LTSP does not LTSP has On Site support while Openthinclient only has remote support 7 1 Discussion The result from this thesis shows that both LTSP and Openthinclient are strong candidates when choosing a thin client solution but are the tests valid The results show that the solutions work when constructed in an isolated environment but when putting the solutions in an environment with existing traffic and normal users the results may vary greatly It does not necessarily mean that the solutions work in a production environment even if they perform well in a laboratory setting Moreover the traffic used during the experiments does not model the behavior of real users since packgen does not provide such functionality The users modeled present thus a static behavior To improve these results to a more realistic environment it would be preferable to conduct these tests in a production environment of a company with normal users using both solutions This does however mean that the results might become very company specific While the results might not be valid in a production environment 1t does show that both of these solutions work adequately in a laboratory se

Thesis Project - Simple search

Contents

Download Pdf Manuals

Related Search

Related Contents