Home

Total cost of ownership analysis for various computing models

1. Inc Total cost of ownership analysis for various computing models 18 Supporting data for performance and sizing calculations 10 11 e Click Finish Create a desktop shortcut to the Q network drive This shortcut s name will be of the form Shortcut to UserX on servername Configure the Visual Test Runtime application so that it will work with all the test scripts a Copy the following five Visual Test dll files into WINDOWS SYSTEM32 e EHelper dll e 6Vtaa dll e VTest60 dll e 6Vtres dll e WebDrive dll b Open a command prompt c Type cd WINDOWS SYSTEM32 and press Enter d For each of the following three dlls type regsvr32 dll filename and press Enter This command registers a dll with the system e EHelper dll e Vtaa dll e WebDrive dll To ensure as consistent a starting point as possible for the performance measurements defragment the hard disk of each PC Using Symantec s Ghost utility make an image of each PC s hard disk This image let us return to a clean and consistent starting point whenever necessary Additional client setup This setup process refers to the creation of the application packages for AppStream and the distribution of those packages to the clients 1 2 3 4 NOOO 9 10 Create a fresh install of Windows XP Professional using all defaults Install AppStream Package Creator Using AppStream Package Creator s Snapshot Package option install
2. Refresh rate Hz Principled Technologies 60 1 Principled Technologies is a registered trademark of Principled Technologies Inc All other product names are the trademarks of their respective owners Disclaimer of Warranties Limitation of Liability 60 Principled Technologies Inc 1007 Slater Rd Suite 250 Durham NC 27703 www principledtechnologies com info principledtechnologies com PRINCIPLED TECHNOLOGIES INC HAS MADE REASONABLE EFFORTS TO ENSURE THE ACCURACY AND VALIDITY OF ITS TESTING HOWEVER PRINCIPLED TECHNOLOGIES INC SPECIFICALLY DISCLAIMS ANY WARRANTY EXPRESSED OR IMPLIED RELATING TO THE TEST RESULTS AND ANALYSIS THEIR ACCURACY COMPLETENESS OR QUALITY INCLUDING ANY IMPLIED WARRANTY OF FITNESS FOR ANY PARTICULAR PURPOSE ALL PERSONS OR ENTITIES RELYING ON THE RESULTS OF ANY TESTING DO SO AT THEIR OWN RISK AND AGREE THAT PRINCIPLED TECHNOLOGIES INC ITS EMPLOYEES AND ITS SUBCONTRACTORS SHALL HAVE NO LIABILITY WHATSOEVER FROM ANY CLAIM OF LOSS OR DAMAGE ON ACCOUNT OF ANY ALLEGED ERROR OR DEFECT IN ANY TESTING PROCEDURE OR RESULT Principled Technologies Inc Total cost of ownership analysis for various computing models Supporting data for performance and sizing calculations 35
3. 2 to quantify the average lost user productivity per day due to platform slowness We looked at two causes of slowness congestion on the access infrastructure servers and in the case of Blade PC desktop computing model the slowness of the blades The next section describes the results Test results For each test script on each test bed we first performed the following script setup steps rebooted in the appropriate order discussed more below the systems in the test bed created a desktop shortcut for the test script created a desktop shortcut for the setup script that prepares the data files for testing created a desktop shortcut for the script that cleans up data files between runs of the script ran the setup script rebooted the systems in the test bed After we finished this setup process for each script we ran that script on that test bed five times for each with a single client then with 2 5 and 10 clients running simultaneously If any test or script failed we discarded that test s results and ran the test again We ran the script that cleans up data files between runs We rebooted the test bed systems the file server and the access infrastructure server if the test bed included one between each run of each test script Figure 3 displays the sum of the execution times for one run of each of the three tests at each of the four client loads for each of the computing models We used these times in calculations of the lost user p
4. BladeSystem bc2000 Blade PC Wyse Winterm V50 thin client General Processor and OS kernel physical core logical UP MP 1P 2C 2L MP 1P 1C 1L UP 1P 1C 1L UP Number of physical processors 1 1 1 Single dual core processors Dual Single Single System Power Management Policy Home Office desk Home Office desk N A CPU Vendor Intel AMD Via Name Core 2 Duo Athlon 64 C3 Model number E6400 2100 Stepping 6 2 Socket type 775 LGA S1 638 Core frequency GHz 2 13 1 2 Front side bus frequency MHz 1 066 800 L1 cache 32 KB 32 KB per core 64 KB 64 KB L2 cache 2 048 KB 512 KB 64 KB Platform Vendor and model number Built in house HP Blade PC bc2000 Wyse Winterm V50 Motherboard model number Intel DQ965GF HP 0A84h Motherboard chipset Intel Q963 ATI RS690 RS690M Motherboard revision number C1 00 Motherboard serial number BQGF64200DXU MXL72316NB BIOS name and version Intel CO96510J 86A 4462 2006 0804 2059 HP 786E5 v01 06 11 BIOS settings Setup default Setup default Principled Technologies Inc Total cost of ownership analysis for various computing models Supporting data for performance and sizing calculations Setup default 33 Clients Hand built standard
5. File server s Figure 8 Terminal Presentation server test bed Principled Technologies Inc Total cost of ownership analysis for various computing models 26 Supporting data for performance and sizing calculations Typically managed rich desktop test bed Figure 9 illustrates the Typically managed rich desktop test bed which includes the following components e Rich Clients hand built Processor 2 13 GHz Intel Core 2 Duo E6400 Memory 1GB DDR2 SRAM Motherboard Intel DQ965GF Disk 160 GB WD Caviar WD1600JS Operating System Windows XP Professional version 2002 with Service Pack 2 00000 e File Server hand built o Processors One 3 4GHz Intel Xeon processor o Memory 2GB DDR2 SRAM o Motherboard HP 382146 405 o Disks One Western Digital WD740ADFD OONLR1 o Operating system Windows Server 2003 R2 Enterprise with Service Pack 2 Rich clients require no additional access infrastructure to Includes the rich access the corporate and or thin clients infrastructure plus any blade PCs or servers the thin clients connect to in order to access the corporate desktop or corporate applications All computations From an access and performance infrastructure point measurements take of view rich clients place directly on the can scale rich clients indefinitely Each concurrent user needs one rich client Includes corporate 1000 Mbps servers that support both thin and rich clients such as database
6. Stop the second stopwatch when the slide viewer finishes displaying the slides that fit in the PowerPoint window 7 Start the third stopwatch and select View Slide Sorter from the PowerPoint menu 8 Stop the third stopwatch when the slider sorter finishes displaying the slides that fit in the PowerPoint window 9 Stop the first stopwatch when Explorer finishes creating the compressed FourFiles zip 10 Close all open applications choosing No if any application asks whether to save changes Principled Technologies Inc Total cost of ownership analysis for various computing models 24 Supporting data for performance and sizing calculations 11 Delete FourFiles zip 12 Empty the Recycle Bin Principled Technologies Inc Total cost of ownership analysis for various computing models Supporting data for performance and sizing calculations 25 Appendix A Overview of test bed configurations In this appendix we present the test beds for each of the compute models we tested Appendix B contains more detailed configuration information Terminal Presentation server test bed Figure 8 illustrates the Terminal Presentation server test bed which includes the following components e Thin clients Wyse Winterm V50 o Processor 1GHz Via C3 o Memory 128MB flash 256 MB DDR RAM o Operating system Wyse Linux V6 e Access infrastructure server hand built o Processors Two 3GHz Intel Xeon 5160 dual core processors Memory 16GB
7. X7DBE motherboard and four 73GB Seagate Cheetah ST373455SS disks and ran Windows Server 2003 R2 Enterprise with Service Pack 2 and other software required by each computing model Appendix A provides high level configuration information on all of the systems used in our test and Appendix B provides greater detail To compare computing model performance we ran three different applications scenarios on each computing model e Acrobat compress This single task scenario tested how quickly the test system was able to open Acrobat and compress a 4 01MB PDF file located on the file server from within the Acrobat application e Excel subtotals This single task scenario tested how quickly Excel could perform the subtotal function on a 1 79MB Excel worksheet located on the file server e Explorer compress and PPT change view This multitasking scenario tested how long it took to compress a 265MB folder located on the file server for the thin clients local for the rich clients while changing views within a 30 4MB Microsoft PowerPoint presentation located on the file server We evaluated the performance of each of the application scenarios with automated test scripts which we developed with IBM s Visual Test 6 5 We hand timed the scenarios to confirm the automated results In this document we focus on the results of the automated scripts because those results are generally more repeatable than hand timings especially in tests involving multip
8. had two servers a file server to hold the test data and another server to hold the configuration applications Because we were testing different Well managed application streaming desktop vPro configurations for the rich clients they too had a file server and an access server Some of the rich client configurations required local installation for the applications the test scripts use The subsequent subsections discuss each test bed and the steps we took to set it up Each of those discussions includes three sections e Test bed specific setup instructions for setting up the access infrastructure server e Instructions for setting up the clients e Additional instructions for client setup Some setup steps were the same for all the test beds while others were specific to each network We note the differences Appendix A describes and diagrams each test bed Setting up the file servers for the test beds We followed this process to initially prepare each file server 1 Install a copy of Microsoft Windows 2003 Server Enterprise Edition R2 2 Apply the following updates from the Microsoft Windows Update site e Windows Server 2003 Service Pack 2 Windows Server 2003 update KB927891 Windows Server 2003 Security Update KB929123 Windows Server 2003 Security Update KB930178 Windows Server 2003 Security Update KB931784 Windows Server 2003 Update KB932168 Windows Server 2003 Security Update KB932168 Windows Server 2003 Security Update KB93356
9. managed rich desktop Well managed OS streaming desktop vPro Blade PC desktop Well managed application streaming desktop vPro Figure 6 Average server CPU during median run for 10 client test Lower is better Typically managed rich desktop and Blade PC desktop platforms do not use access infrastructure servers so their results are not applicable Principled Technologies Inc Total cost of ownership analysis for various computing models 6 Supporting data for performance and sizing calculations Client factor All of our test clients simultaneously performed server intensive tasks To account for the fact that only a percentage of any real life group of clients would be executing server intensive tasks at any given time we assigned a client factor the number of actual users each test client in our tests represented Our default client factor was 10 reflecting an industry standard range of 8 to 10 Estimated number of real clients The product of the raw client count and the client factor was the estimated number of real clients the computing model could support if all computing models executed the tasks in the same amount of time Because this was not the case we used one additional factor to account for the variation in speed Platform speed normalized to rich client For the last factor we divided the task completion time on the rich client platform by the task completion time on the shared server platform Poorer
10. models access infrastructure servers to host the desktop and any running applications e Application infrastructure that includes one or more file servers as well as application servers that provide email database network services DNS Active Directory and Web services e Client management infrastructure that automates client management software Figure 2 defines the three infrastructures within a general compute model enterprise view and Appendix A lists the test bed components and illustrates the computing models for each of the five computing models Each test bed included client systems and a file server to store the test files The thin clients do not have disks so they had an access infrastructure server to host the desktop and any running applications Because client management would not affect the access infrastructure server performance central to this study we did not include client management servers in our test bed other than the application streaming server Likewise we omit from our test bed the application servers other than a file server that enterprise infrastructures typically include Principled Technologies Inc Total cost of ownership analysis for various computing models 3 Supporting data for performance and sizing calculations The analysis we document in this paper had two goals 1 to estimate the number of clients that an access infrastructure server can realistically support on computing models that use them and
11. of the scripts one for each user These folders contain all of the files the scripts need to execute Each of the three script folders names identifies the test script Each of those folders contains a folder named SC1 Each SC1 folders contains the same three subfolders a Content all the test files the script uses b Results initially empty but will contain the results file the script creates c Scripts the script s source and executable files 8 To ensure as consistent a starting point as possible for the performance measurements defragment the hard disk 9 Using Symantec s Ghost utility make an image of the hard disk This image let us return to a clean and consistent starting point whenever necessary Setting up the Terminal Presentation server test bed We set up the Terminal Presentation server test bed by arranging the 10 Wyse Winterm V50 thin clients in a network with a 100Mbps switch We connected the file server and access infrastructure server to each other through a 1Gbps switch We then connected the two switches to each other using a 1Gbps uplink port Install a copy of Microsoft Windows 2003 Server Enterprise Edition R2 2 Apply the following updates from the Microsoft Windows Update site e Windows Server 2003 Service Pack 2 Windows Server 2003 update KB927891 Windows Server 2003 Security Update KB929123 Windows Server 2003 Security Update KB930178 Windows Server 2003 Security Update KB931784 Windows Se
12. on the file server The script The script for the scenario performs the following tasks Open Sales2002a1 xls We did not time that task because we focused on a single function Start the Excel timer and perform the subtotal function Stop the Excel timer when Excel finishes calculating the subtotals Close Excel We did not time that task because we focused on a single function The manual process To execute the test follow these instructions You will need one stopwatch 1 Reboot the system 2 Open Sales2002a1 xls 3 Select Data Subtotals 4 Fill in the Subtotal dialog as follows e Ateach change in Size e Use function Average e Add subtotal to Quantity e Uncheck Replace current subtotals 5 Start stopwatch and press OK 6 Stop stopwatch when the Inserting Subtotals progress bar goes away and the status bar says Ready 7 Close Excel choosing No when Excel asks whether you want to save Single task scenario Compressing a PDF from within Adobe Acrobat The application involved e Adobe Acrobat 8 0 Professional The data file involved e Computing pdf a 4 01MB PDF file on the file server The script The script for the scenario performs the following tasks Open Computing pdf We did not time that task because we focused on a single function Start the Acrobat timer and tell Acrobat to compress the PDF Stop the Acrobat timer when the Conversion Warning dialog displays Close the Conversion Warning w
13. performance on the shared server platform than on the rich client platform indicated that the server had too many clients because we treated the rich client platform as the baseline for performance Put differently normalizing to the rich client s performance estimated the number of clients the server could handle while providing the rich client s level of performance The server for the OS streaming platform was not constrained by the processor as it was on all of the other alternative computing models To account for this we used three special tests we describe below We then used the minimum calculated number from these formulas to estimate the number of clients per OS streaming server 1 Calculating clients per server during a disk intensive task In this case the task was compressing a large file located on the OS streamed disk We used the same formula as above during this disk intensive task number of clients number of clients rich client time percentage CPU utilization each client shared platform time represents 2 Measuring network utilization during a 10 client boot storm number of real number of clients x clients x _OS streaming single client boot time percentage network each test client OS streaming 10 client boot time utilization represents raw client count client factor OS streaming 10 client boot time normalized to the OS streaming estimated number of real clients single client boot time Raw clien
14. servers directory servers Web servers middleware servers print servers file servers etc For our tests we use a file server as representative While File server s all client types strain the application infrastructure the load does not differ significantly by client type Figure 11 Blade PC desktop test bed Principled Technologies Inc Total cost of ownership analysis for various computing models 29 Supporting data for performance and sizing calculations Well managed application streaming desktop vPro test bed Figure 12 illustrates the Well managed application streaming desktop vPro test bed which includes the following components e Rich Clients hand built Processor 2 13 GHz Intel Core 2 Duo E6400 Memory 1GB DDR2 SRAM Motherboard Intel DQ965GF Disk 160 GB WD Caviar WD1600JS Operating System Windows XP Professional version 2002 with Service Pack 2 Application streaming software AppStream Technology Windows Edition 5 2 1 Client 000000 e AppStream Server hand built Processors Two 3GHz Intel Xeon 5160 dual core processors Memory 16GB DDR2 SRAM Motherboard Supermicro X7DBE Disks Four 73GB Seagate Cheetah ST373455SS disks Operating system Windows Server 2003 R2 Enterprise with Service Pack 2 Application streaming software AppStream Technology Windows Edition 5 2 1 Server O 00000 e File Server hand built o Processors One 3 4GHz Intel Xeon processor o Memory 2GB DDR2 S
15. the case we used one additional factor to account for the variation in speed OS streaming 10 client test time normalized to the OS streaming single client test time For the last factor we divided the test completion time of a single client by the 10 client test completion time Normalizing to the single client performance estimated the number of clients the server could handle while providing the single client s level of performance Figure 7 shows the results of these calculations Principled Technologies Inc Total cost of ownership analysis for various computing models 8 Supporting data for performance and sizing calculations Terminal Presentation server 56 57 67 51 58 27 Typically managed rich desktop n a n a n a Well managed OS streaming desktop vPro 15036 25 9548 80 1239 27 Blade PC desktop n a n a n a Well managed Application 399 00 21153 40 628 27 streaming desktop vPro Figure 7 Estimated number of users per server determined by the worst case number of users per server for the three or four tests These estimates did not apply to Typically managed rich desktop or Blade PC desktop platforms none of which require access infrastructure servers Test bed setup We performed the initial setup of the shared file server the same way on all the test beds The first subsection below outlines that process The thin clients do not have disks so they
16. xlcpq 5 88 4 333 Type Integrated Integrated Memory size 16 MB 8 MB shared Resolution 1 280 x 1 024 1 280 x 1 024 Driver Microsoft 5 2 3790 0 ATI 5 10 2600 6014 Network card subsystem Vendor and model number Intel PRO 1000 EB Network Connection HP NC7782 Gigabit Server Adapter Type Integrated Integrated Driver Intel 9 7 34 0 HP 10 39 0 0 Additional card information HP NC7782 Gigabit Server Adapter 2 Additional card type Integrated Additional card driver HP 10 39 0 0 Optical drive Vendor and model number Type Interface Dual single layer USB ports MATSHITA SR 8178 DVD ROM SATA Dual HLDS GCR 8240N CD ROM SATA Single Number of ports 4 Type of ports USB1 1 USB2 0 3 USB 1 0 1 USB 2 0 RAID Type RAID 0 Size 274 GB Driver version Intel Logic Corp 1 23 0 32 Principled Technologies Inc Total cost of ownership analysis for various computing models 32 Supporting data for performance and sizing calculations Servers Supermicro X7DB8 access infrastructure server HP ProLiant DL360 file server RAID devices Vendor and model number Intel RAID Controller SRSCSAS144E Firmware version 1 03 00 0211 BIOS version 1 03 08_02 Host interface Clients Clients PCI Express Hand built standard rich client HP
17. 5 15 3 3 3 11 Size 16 382 MB 2 048 MB Number of RAM modules 16 x 1 024 MB 2 x 1 024 MB Chip organization Double sided Double sided Channel Dual Single Hard disk Vendor and model number Seagate ST373455SS Western Digital WD740ADFD OONLR1 Number of disks in system 4 1 Size 73 GB 74 GB Buffer size 16 MB 16 MB RPM 15 000 10 000 Type SAS SATA Controller Intel RAID Controller SRSCSAS144E Intel 6300ESB ICH S Controller driver Intel Logic Corp 1 23 0 32 Microsoft 5 2 3790 0 Principled Technologies Inc Total cost of ownership analysis for various computing models Supporting data for performance and sizing calculations 31 Servers Supermicro X7DB8 access infrastructure server HP ProLiant DL360 file server Operating system Windows Name Microsoft Windows Server 2003 R2 Enterprise Microsoft Windows Server 2003 R2 Enterprise Build number 3790 3790 Service Pack SP2 SP2 Microsoft Windows update date 07 03 2007 07 03 2007 File system NTFS NTFS Kernel ACPI Multiprocessor PC ACPI Multiprocessor PC Language English English Microsoft DirectX version 9 0c 9 0c Graphics Vendor and model number ATI ES1000 ATI RAGE XL Chipset ES1000 ATI RAGE XL BIOS version 01 00 GR
18. 6 Windows Server 2003 Security Update KB935839 Windows Server 2003 Security Update KB935840 Windows Server 2003 Security Update KB924667 v2 Windows Server 2003 Security Update KB925902 3 Turn off Windows Service Pack 2 Security Center Pop up Alerts Doing so prevents such alerts from occurring during testing and affecting results a Open the Control Panel b Choose Security Center c Click Change the way Security Center Alerts me on the left Principled Technologies Inc Total cost of ownership analysis for various computing models Supporting data for performance and sizing calculations d Uncheck Firewall Automatic Updates and Virus Protection 4 Turn off Windows Automatic Updates Doing so prevents such updates from occurring during testing and affecting results a Open the Control Panel b Choose Automatic Updates c Select Turn off Automatic Updates 5 Turn off System Restore Doing so prevents such events from occurring during testing and affecting results a Open the Control Panel b Choose System c Choose the System Restore tab d Select Turn off System Restore on all drives 6 Setup the script folders for each test system PC or thin client a Create a folder named User Files b Within that folder create 10 folders named User1 through User10 These folders will hold the individual script directories 7 Copy the three script folders into each of these directories You will end up with 10 copies
19. 9 0c 9 0c Graphics Vendor and model number Intel GMA 3000 ATI Type Integrated Integrated Integrated Chipset Intel GMA 3000 BIOS version 1348 Memory size 128 MB Resolution 1 280 x 1 024 1 280 x 1 024 1 280 x 1 024 Driver Intel 6 14 10 4618 Sound card subsystem Vendor and model number Driver Sigma Tel High Definition Audio CODEC Sigma Tel 5 10 5067 0 Principled Technologies Inc Total cost of ownership analysis for various computing models Supporting data for performance and sizing calculations Integrated 34 Clients Hand built standard rich client HP BladeSystem bc2000 Blade PC Wyse Winterm V50 thin client Ethernet Vendor and model number Intel 82566DM Gigabit Network Connection Broadcom NetLink Fast Ethernet 10 100 Base T Fast Ethernet twisted pair Driver Intel 9 4 17 0 Broadcom 10 24 0 0 RJ 45 Optical drive s Vendor and model number Type Interface Dual single layer USB ports TSSTcorp SH S183L DVD CD ROM RW SATA Dual layer N A N A N A N A N A N A N A N A Number 10 2 only with adapter 3 Type 2 0 2 0 2 0 IEEE 1394 Number 0 0 Monitor Type ViewSonic Optiquest Q7 ViewSonic Optiquest Q7 ViewSonic Optiquest Q7 Screen size 17 17 17
20. Center Click Change the way Security Center Alerts me on the left Uncheck Firewall Automatic Updates and Virus Protection 6 Turn off Windows Automatic Updates Doing so prevents such updates from occurring during testing and affecting results a Open the Control Panel b Choose Automatic Updates c Select Turn off Automatic Updates 7 Turn off System Restore Doing so prevents such events from occurring during testing and affecting results a Open the Control Panel b Choose System c Choose the System Restore tab d Select Turn off System Restore on all drives 8 Change the IP address to 192 168 1 5 X where X is the client number of the blade PC We numbered the blade PCs 1 through 10 with the last address being 192 168 1 60 Click Start gt Control Panel gt Select Network Connections Select Local Area Connection In the Local Area Connection Status dialog select Properties In the Local Area Connection Properties dialog select Properties Select the Use the following IP address radio button Enter the IP Address Select OK on the top two open dialog boxes Select Close to close the third dialog box ap the network drive Q to servername fileserver user files user X where X is as above Right click My Computer Select Map Network Drive Select Q from the Drive drop down menu Type servername fileserver user files user X where X is as above as the Folder name Click Finish 10 Create a desktop shortcut to t
21. DDR2 SRAM Motherboard Supermicro X7DBE Disks Four 73GB Seagate Cheetah ST373455SS disks Operating system Windows Server 2003 R2 Enterprise with Service Pack 2 Thin client management software Citrix Presentation Server 4 5 o0o0000 e File Server hand built o Processors One 3 4GHz Intel Xeon processor o Memory 2GB DDR2 SRAM o Motherboard HP 382146 405 o Disks One Western Digital WD740ADFD OONLR1 disk o Operating system Windows Server 2003 R2 Enterprise with Service Pack 2 Thin clients require Includes the rich and or thin an access clients plus any blade PCs or infrastructure such servers the thin clients connect as Citrix to in order to access the Presentation Server corporate desktop or corporate to access the Wyse thin client applications corporate infrastructure 1 through 10 We calculate the All computations number of potential and performance thin clients per measurements take server based on place directly on the metrics from our Citrix Presentation P script and server Server performance testing Citrix Presentation Server Includes corporate servers that SY support both thin and rich 7 clients such as database 1000 Mbps servers directory servers Web servers middleware servers print servers file servers etc For our tests we use a file server as representative While all client types strain the application infrastructure the i load does not differ significantly by client type
22. Follow these steps to ready all servers and clients for testing 1 Power on the file server 2 Power on the AppStream server 3 When the servers are fully active power on the PCs you are testing Rebooting the Well managed application streaming desktop vPro test bed systems Follow these steps to ready the servers and PCs for testing Click Start gt Shut Down on each active PC Select Shut Down from the menu on each active PC Log on to the file server as administrator Click Start gt Shut Down Make sure that in the list box under What do you want the computer to do you have selected Restart Type an explanatory comment in the comment box Press Enter You may see a warning window that says shutting down will close the remote connections and that asks if you want to end them Click Yes and the file server will restart 9 Log onto the AppStream server as administrator 10 Click Start gt Shut Down 11 Make sure that in the list box under What do you want the computer to do you have selected Restart 12 Press Enter 13 You may see a warning window that says shutting down will close the remote connections and that asks if you want to end them Click Yes and the server will restart 14 When the servers are fully active power on the PCs you are testing o NON OIL Oo D a Application scenarios Descriptions and instructions for running manually This section summarizes the script for each application scenario and exp
23. Microsoft Office 2003 After Package Creator has finished creating the Office package copy the package onto the access infrastructure server Repeat steps 1 and 2 Using AppStream Package Creator s Snapshot Package option install Adobe Acrobat 8 0 Professional After Package Creator has finished creating the Acrobat package copy the package onto the access infrastructure server Log on to the AppStream interface and select the Well managed application streaming desktop vPro server Load the two packages you created earlier Go to the AppStream users and give them permission to access the two created packages Running the tests Setting up the servers and clients is much more complicated than the testing process When you have correctly set up the clients they will automatically connect to the server s Executing the tests on a particular type of client involves three relatively straightforward phases 1 2 Getting the systems ready to go In this phase you make sure all the systems in the test bed are on appropriately connected i e clients are connected to the file server and ready for testing Setting up the test script you want to run Each script has a setup script that you must run once on each client before testing on that client with that script The setup script makes sure the data files are ready the application windows are where the test script expects to find them and so on Running the test scripts and recordin
24. P address radio button Enter the IP Address Select OK on the top two open dialog boxes Select Close to close the third dialog box 5 Configure each thin client to have a display resolution of 1 280 x 1 024 and 24 bit color depth 6 Create a remote desktop connection with the Citrix server a Click Start gt Connection Manager b In the Connection manager window click Add c In the first window select ICA and click Next d Under the Network tab enter the following i Inthe Description box Citrix sa oa290T D Principled Technologies Inc Total cost of ownership analysis for various computing models 11 Supporting data for performance and sizing calculations ii In the Server box 192 168 1 250 e Under the Window tab set the Window Colors to 16 Million and the Window Size to Full screen f Under the Login tab enter the following i In the User Name box user X where X is as above ii For the password password iii For the Domain server Double click the Citrix icon to connect to the Citrix server Map the network drive Q to servername fileserver user files user X where X is as above a Right click My Computer b Select Map Network Drive c Select Q from the Drive drop down menu d Type servername fileserver user files user X where X is as above as the Folder name e Click Finish Create a desktop shortcut to the Q network drive This shortcut s name will be of the form Shortcut to UserX o
25. RAM o Motherboard HP 382146 405 o Disks One Western Digital WD740ADFD OONLR1 o Operating system Windows Server 2003 R2 Enterprise with Service Pack 2 We measure performance on the AppStream server We calculate the maximum number of concurrent users based on scripts and server performance data The application streaming server is a management server that downloads applications to the rich clients on demand AppStream ae R A Rich client 1 through 10 File server s Includes the rich and or thin clients plus any blade PCs or servers the thin clients connect to in order to access the corporate desktop or corporate applications All computations and most performance measurements take place directly on the rich clients AppStream clients require no additional access infrastructure to access the corporate infrastructure Includes corporate servers that support both thin and rich clients such as database servers directory servers Web servers middleware servers print servers file servers etc For our tests we use a file server as representative While all client types strain the application infrastructure the load does not differ significantly by client type Figure 12 Well managed application streaming desktop vPro test bed Principled Technologies Inc Total cost of ownership analysis for various computing models 30 Supporting data for performance and sizing calculations Appendi
26. SUPPORTING DATA OCTOBER 2007 Total cost of ownership analysis for various computing models Supporting data for performance and sizing calculations Executive summary Intel Corporation Intel commissioned Principled Technologies PT to compare the performance of five computing models on test beds with varying numbers of active clients each running the same typical office application tasks We measured the response time users would experience on each system while performing common office functions with leading Microsoft Office and Adobe Acrobat applications We set up test beds to test the following computing models e Terminal Presentation server thin clients administered using Citrix Presentation Server 4 5 e Typically managed rich desktop rich clients e Well managed OS streaming desktop vPro rich clients with Intel vPro technology with the OS streamed from the server and Ardence Evaluation Client 4 1 administered using Ardence Evaluation Server 4 1 OS e Blade PC desktop thin clients connected to Blade PCs e Well managed application streaming desktop vPro rich clients with Intel vPro technology running AppStream Technology Windows Edition 5 2 1 client software with applications streamed using AppStream Technology Windows Edition 5 2 1 server software All rich clients were hand built rich clients Intel Desktop Board DQ965GF 2 13 GHz Intel Core 2 Duo E6400 1GB DDR2 SRAM running Windows XP Professional version 2002 wi
27. We then connected the two switches to each other using a 1Gbps uplink port Setting up the access infrastructure server in the Well managed OS streaming desktop vPro test bed 1 Install a copy of Microsoft Windows 2003 Server Enterprise Edition R2 2 Apply the following updates from the Microsoft Windows Update site e Windows Server 2003 Service Pack 2 Windows Server 2003 update KB927891 Windows Server 2003 Security Update KB929123 Windows Server 2003 Security Update KB930178 Windows Server 2003 Security Update KB931784 Windows Server 2003 Update KB932168 Windows Server 2003 Security Update KB932168 Windows Server 2003 Security Update KB933566 Windows Server 2003 Security Update KB935839 Windows Server 2003 Security Update KB935840 Windows Server 2003 Security Update KB924667 v2 Principled Technologies Inc Total cost of ownership analysis for various computing models 13 Supporting data for performance and sizing calculations e Windows Server 2003 Security Update KB925902 3 Turn off Windows Service Pack 2 Security Center Pop up Alerts Doing so prevents such alerts from occurring during testing and affecting results a Open the Control Panel b Choose Security Center c Click Change the way Security Center Alerts me on the left d Uncheck Firewall Automatic Updates and Virus Protection 4 Turn off Windows Automatic Updates Doing so prevents such updates from occurring during testing and affecting
28. anup program If so the Script folder will contain a third file named SC1 Prep pc6 or SC1cleanup pc6 respectively If either file exists create a desktop shortcut to it 7 Run SC1 Setup Phase 3 is largely the same regardless of the type of client Once you have finished the above script setup phase do the following for each test you want to run 1 Reboot all the servers and the clients you will be using in the test This process varies by client type we outline it for each client test bed below 2 Wait 10 seconds after the Windows hourglass has disappeared on all the clients to ensure a consistent starting state 3 On each client you want to test if there is a shortcut to SC1 Prep or SC1cleanup do the following a Double click that shortcut b Wait until you see a confirmation window indicating that prep has completed or in the case of SC1cleanup wait 30 seconds 4 Start the script at the same time on all the clients you are testing by clicking the Shortcut to SC1main and pressing Enter on each client 5 When the test completes record the results of each client As we discussed at the beginning of the Test methodology section we ran each script five times on each test configuration of each network e g five times with one active PC five times with two active PCs and so on In the following three subsections we detail the first phase for each of the three types of test beds Testing the Terminal Presentation server comput
29. ation streaming desktop vPro test bed We set up the Well managed application streaming desktop vPro test bed by arranging the 10 desktop systems in a network with a 100Mbps switch We connected the file server and access infrastructure server to each other through a 1Gbps switch We then connected the two switches to each other using a 1Gbps uplink port Setting up the access infrastructure server in the Well managed application streaming desktop vPro test bed Install a copy of Microsoft Windows 2003 Server Enterprise Edition R2 2 Apply the following updates from the Microsoft Windows Update site e Windows Server 2003 Service Pack 2 Windows Server 2003 update KB927891 Windows Server 2003 Security Update KB929123 Windows Server 2003 Security Update KB930178 Windows Server 2003 Security Update KB931784 Windows Server 2003 Update KB932168 Windows Server 2003 Security Update KB932168 Windows Server 2003 Security Update KB933566 Windows Server 2003 Security Update KB935839 Windows Server 2003 Security Update KB935840 Windows Server 2003 Security Update KB924667 v2 Windows Server 2003 Security Update KB925902 3 Turn off Windows Service Pack 2 Security Center Pop up Alerts Doing so prevents such alerts from occurring during testing and affecting results a Open the Control Panel b Choose Security Center Principled Technologies Inc Total cost of ownership analysis for various computing models Supporting data for p
30. be used only once at the beginning of the network setup Only one client is involved so do not repeat these steps on the other systems 1 Install Windows XP Professional 2 Apply all XP critical updates as of July 10 2007 from the Microsoft Windows Update Web site including Windows XP SP2 Do not install any of the optional updates 3 Install Microsoft Office 2003 4 Apply all Office 2003 updates as of July 10 2007 from the Microsoft Office Update Web site 5 Turn off Windows Service Pack 2 Security Center Pop up Alerts Doing so prevents such alerts from occurring during testing and affecting results a Open the Control Panel b Choose Security Center c Click Change the way Security Center Alerts me on the left d Uncheck Firewall Automatic Updates and Virus Protection 6 Turn off Windows Automatic Updates Doing so prevents such updates from occurring during testing and affecting results a Open the Control Panel b Choose Automatic Updates c Select Turn off Automatic Updates 7 Turn off System Restore Doing so prevents such events from occurring during testing and affecting results Principled Technologies Inc Total cost of ownership analysis for various computing models 14 Supporting data for performance and sizing calculations a Open the Control Panel b Choose System c Choose the System Restore tab d Select Turn off System Restore on all drives Map the network drive Q to servername fileserver u
31. ct Properties In the Local Area Connection Properties dialog select Properties Select the Use the following IP address radio button Enter the IP Address Select OK on the top two open dialog boxes Select Close to close the third dialog box 5 Configure each thin client to have a display resolution of 1 280 x 1 024 and 24 bit color depth 6 Create a remote desktop connection with the Blade PC Enclosure a Click Start gt Connection Manager b Inthe Connection manager window click Add c In the first window select RDP and click Next d Under the Network tab enter the following i Inthe Description box RDP Connection ii Inthe Server box 192 168 1 5 X where X is as above Under the Window tab set the Window Colors to 16 Million and the Window Size to Full screen Under the Login tab enter the following i Inthe User Name box administrator ii For the password password 7 Double click the RDP Connection icon to connect to the VM 8 Map the network drive Q to servername fileserver user files user X where X is as above a Right click My Computer b Select Map Network Drive c Select Q from the Drive drop down menu d Type servername fileserver user files user X where X is as above as the Folder name e Click Finish sa gt o9a0TD o 9 Create a desktop shortcut to the Q network drive This shortcuts name will be of the form Shortcut to UserX on servername Setting up the Well managed applic
32. d the password password 9 Install Citrix Presentation Server using all defaults 10 Set up USER1 through USER10 so each account has Citrix user and remote desktop permissions 11 Configure the Visual Test Runtime application so that it will work with all the test scripts a Copy the following five Visual Test dll files into WINDOWS SYSTEM32 e EHelper dll e 6Vtaa dll e VTest60 dll e 6Vtres dll e WebDrive dll b Open a command prompt c Type cd WINDOWS SYSTEM32 and press Enter d For each of the following three dlls type regsvr32 dll filename and press Enter This command registers a dll with the system e EHelper dll e Vtaa dll e WebDrive dll 12 To ensure as consistent a starting point as possible for the performance measurements defragment the hard disk 13 Using Symantec s Ghost utility make an image of the hard disk This image let us return to a clean and consistent starting point whenever necessary 14 Install Adobe Acrobat 8 0 Professional Setting up the thin clients 1 Boot the file server 2 Boot the Citrix server 3 Boot the thin client 4 Change the IP address of the thin client to 192 168 1 5 X where X is the number of the client Click Start gt Control Panel gt Select Network Connections Select Local Area Connection In the Local Area Connection Status dialog select Properties In the Local Area Connection Properties dialog select Properties Select the Use the following I
33. dence Server Administrator 17 Change the prime client to boot from the vDisk 18 Boot up all the other clients 19 At the prompt to add the clients to Ardence choose the newly created vDisk and add them Setting up the Blade PC desktop test bed We set up the Blade PC desktop test bed by arranging the 10 Wyse Winterm V50 thin clients in a network with a 100Mbps switch We hosted the HP bc2000 blades in a Blade PC Enclosure We connected the file server and Blade PC Enclosure to each other through a 1Gbps switch We then connected the two switches to each other using a 1Gbps uplink port Setting up the access infrastructure server in the Blade PC desktop test bed The Blade PC desktop computing model does not use an access infrastructure server Setting up the Blade PC desktop clients Perform these steps on each blade PC 1 Install Windows XP Professional 2 Apply all XP updates as of July 10 2007 from the Microsoft Windows Update Web site including Windows XP SP2 3 Install Microsoft Office 2003 4 Apply all Office 2003 updates as of July 10 2007 from the Microsoft Office Update Web site 5 Turn off Windows Service Pack 2 Security Center Pop up Alerts Doing so prevents such alerts from occurring during testing and affecting results Principled Technologies Inc Total cost of ownership analysis for various computing models 15 Supporting data for performance and sizing calculations Open the Control Panel Choose Security
34. e shared server penalty in productivity Figure 4 shows the results of this calculation Terminal Presentation server Typically managed rich desktop Well managed OS streaming desktop vPro Blade PC desktop Well managed application streaming desktop vPro Figure 4 Script execution times for 10 users and shared server penalty all times in seconds lower is better We estimated that the typical eight hour workday contains at least the following four periods of peak usage when more than 10 users per server are simultaneously executing server intensive tasks and would therefore encounter delays equal to or greater than those we report with 10 clients e at the beginning of the workday e before lunch e after lunch e at the end of the workday To calculate a conservative estimate for total daily penalty per user per eight hour workday we multiplied the shared server penalty by four In the case of the Blade PC desktop however the script execution times were significantly slower than rich clients at even one concurrent user Therefore we estimate that 10 times during the eight hour workday a typical user is attempting compute intensive tasks We then multiply the shared server penalty for Blade PC desktop by 10 to get the total daily penalty for an eight hour workday Principled Technologies Inc Total cost of ownership analysis for various computing models 5 Supporting data for performance and sizing calc
35. ecution preparation steps specific to the Typically managed rich desktop test bed Getting the Typically managed rich desktop test bed systems ready to go Follow these steps to ready the file server and PCs for testing 1 Power on the file server 2 When the server is fully active power on the PCs you are testing Rebooting the Typically managed rich desktop test bed systems Follow these steps to reboot the file server and PCs for testing 1 Click Start gt Shut Down on each active PC Select Shut Down from the menu on each active PC Log on to the file server as administrator Click Start gt Shut Down Make sure that in the list box under What do you want the computer to do you have selected Restart Type an explanatory comment in the comment box Press Enter You may see a warning window that says shutting down will close the remote connections and asks if you want to end them Click Yes and the file server will restart 9 When the server is fully active power on the PCs you are testing o DLO TE O N Testing the Well managed OS streaming desktop vPro computing model This section provides the test execution preparation steps specific to the Well managed OS streaming desktop vPro PCs Getting the Well managed OS streaming desktop vPro test bed systems ready to go Follow these steps to ready the file server Ardence server and the Ardence PCs for testing 1 Power on the file server 2 Power on the Ardence server 3 When the
36. erformance and sizing calculations c Click Change the way Security Center Alerts me on the left d Uncheck Firewall Automatic Updates and Virus Protection 4 Turn off Windows Automatic Updates Doing so prevents such updates from occurring during testing and affecting results a Open the Control Panel b Choose Automatic Updates c Select Turn off Automatic Updates 5 Turn off System Restore Doing so prevents such events from occurring during testing and affecting results a Open the Control Panel b Choose System c Choose the System Restore tab d Select Turn off System Restore on all drives Install AppStream Server on the server keeping all defaults Create a user for the clients of AppStream To ensure as consistent a starting point as possible for the performance measurements defragment the hard disk 9 Using Symantec s Ghost utility make an image of the hard disk This image let us return to a clean and consistent starting point whenever necessary oN Setting up the Well managed application streaming desktop vPro clients Repeat the steps on each of the 10 clients 1 Install Windows XP Professional 2 Apply all XP critical updates as of July 10 2007 from the Microsoft Windows Update Web site including 3 Turn off Windows Service Pack 2 Security Center Pop up Alerts Doing so prevents such alerts from occurring during testing and affecting results a Open the Control Panel b Choose Security Center c Clic
37. following tasks e Open Windows Explorer We did not time this task because it occurs outside the multitasking section of the script Navigate to FourFiles Start the timer for Explorer Compress and start compressing FourFiles Start the PowerPoint Open timer and double click the Content ppt desktop shortcut Stop the PowerPoint Open timer when the bottom slide in the slide viewer loads Start the PowerPoint Change View timer and select View Slide Sorter Stop the PowerPoint Change View timer when the last slide loads in the Slide Sorter view Stop the Explorer Compress timer when Explorer finishes compressing the file Close all open applications We did not time these tasks because they occur outside the multitasking section of the script The manual process First prepare each system by following these steps once 1 Open Windows Explorer 2 Ensure that it allows enough space on the right to allow double clicking a desktop shortcut there 3 Create a desktop shortcut to Content ppt Place it on the right side of the desktop to the right of the Explorer window To execute the test follow these instructions You will need three stopwatches to time three different tasks 1 Reboot the system 2 Open Windows Explorer 3 Navigate to the folder that contains FourFiles 4 Start the first stopwatch and click Send to Compressed zipped folder 5 Start the second stopwatch and double click the Content ppt desktop shortcut 6
38. g models 4 Supporting data for performance and sizing calculations frequently work in medium to large files Typical applications include Microsoft Office and complex line of business applications such as accounting and resource planning applications Our three tests include processor intensive operations in Microsoft Office and Adobe Acrobat applications that are typical of those that knowledge workers use day to day The results for our example enterprise should be relevant to other enterprises selecting a computing model for a group of predominantly knowledge workers An enterprise with all light or transactional users would see less lost productivity and an organization with more heavy or power users would likely need more servers and experience additional lost productivity Shared server penalty and lost productivity per user per day When multiple users are simultaneously executing server intensive tasks some or all users may have to wait because the server is unable to process all their work quickly enough To quantify the amount of time users lose to waiting during these instances we followed this process 1 Sum the total number of seconds it took each client to complete our three tests on the computing model in question when 10 users were running at once See the 10 client column in Figure 3 or the total times column in Figure 4 2 Subtract the sum of the times the rich clients required to execute the same tasks The result is th
39. g results You must reboot the test bed systems before each run of each test script and start the test script at the same time on all the clients under test Phase 1 varies for each test bed We detail it below in the sections on the test beds In all of these discussions we assume you have already completed the setup process we outlined earlier We also assume any client systems you do not want to include in a test will not be on Principled Technologies Inc Total cost of ownership analysis for various computing models 19 Supporting data for performance and sizing calculations Phase 2 is the same regardless of the type of client you are testing Once you have readied all the systems to go and are working on a client follow this process to prepare the client to run a test script 1 Double click the desktop shortcut Shortcut to UserX at servername where X is the number of the client and servername is the name you gave the file server 2 You will see four folders one for each script Open the folder that contains the script you are testing 3 Inside that folder you will find a folder named SC1 Double click that folder 4 You will see three folders Content Results and Scripts The Scripts folder contains the individual script files Double click the Scripts folder 5 Inthe Scripts folder find the files SC1 Setup pc6 and SC1main pc6 Create desktop shortcuts to each of them 6 Some scripts require an additional preparation or cle
40. he Q network drive This shortcut s name will be of the form Shortcut to UserX on servername 11 Configure the Visual Test Runtime application so that it will work with all the test scripts a Copy the following five Visual Test dll files into WINDOWS SYSTEM32 apog oonooowezra oanan e EHelper dll e Vtaa dll e VTest60 dll e Vtres dll e WebDrive dll b Open a command prompt c Type cd WINDOWS SYSTEM32 and press Enter d For each of the following three dlls type regsvr32 dll filename and press Enter This command registers a dll with the system e EHelper dll e Vtaa dll e WebDrive dll 12 Change the blade PC to allow remote connections to the desktop a Right click on My Computer and select Properties b Click on the Remote tab in the System Properties window c Under Remote Desktop select the option allowing remote desktop connections 13 Install Adobe Acrobat 8 0 Professional Additional Blade PC desktop client setup 1 Boot the file server 2 Boot the Blade PC Enclosure Principled Technologies Inc Total cost of ownership analysis for various computing models 16 Supporting data for performance and sizing calculations 3 Boot the thin client 4 Change the IP address of the thin client to 192 168 1 6 X where X is the number of the client Click Start gt Control Panel gt Select Network Connections Select Local Area Connection In the Local Area Connection Status dialog sele
41. igital WD740ADFD OONLR1 o Operating system Windows Server 2003 R2 Enterprise with Service Pack 2 Includes the rich and or thin clients plus any blade PCs or servers the thin clients N connect to in order to access the corporate desktop or l X corporate applications S SS P S The rich clients boot from r9 the Ardence OS streaming Ardence OS i 7 r gt server using PXE and a All computations and Rich client 1 through 10 streaming server custom disk driver We most performances capture performance f metrics using Perfmon on place directly on the the Ardence server and rich client However k 7 the Ardence server munple scripts on the remotely hosts the OS storage We calculate the number of potential OS streaming rich clients per server based on script and server performance metrics Includes corporate servers that support both thin and rich clients such as database servers directory servers Web servers middleware servers print servers file servers etc For our tests we use a file server as representative While all client types strain the application infrastructure the load does not differ significantly by client type File server s Figure 10 Well managed OS streaming desktop vPro test bed Principled Technologies Inc Total cost of ownership analysis for various computing models Supporting data for performance and sizing calculations Blade PC desktop test bed Figure 11 illustra
42. ileserver user files user X where X is as above as the Folder name e Click Finish 10 Create a desktop shortcut to the Q network drive This shortcuts name will be of the form Shortcut to UserX on servername 11 Configure the Visual Test Runtime application so that it will work with all the test scripts a Copy the following five Visual Test dll files into WINDOWS SYSTEM32 aooMzsa gt 9a9 e EHelper dll e Vtaa dll e VTest60 dll e Vtres dll e WebDrive dll b Open a command prompt c Type cd WINDOWS SYSTEM32 and press Enter d For each of the following three dlls type regsvr32 dll filename and press Enter This command registers a dll with the system e EHelper dll e Vtaa dll e WebDrive dll 12 To ensure as consistent a starting point as possible for the performance measurements defragment the hard disk of each PC 13 Using Symantec s Ghost utility make an image of each PC s hard disk This image let us return to a clean and consistent starting point whenever necessary 14 Install Adobe Acrobat 8 0 Professional Additional client setup The PC test bed does not require any additional client setup Setting up the Well managed OS streaming desktop vPro test bed We set up the Well managed OS streaming desktop vPro test bed by arranging the 10 desktop systems in a network with a 100Mbps switch We connected the file server and access infrastructure server to each other through a 1Gbps switch
43. indow Close Acrobat Delete Compress pdf the file the script just created We did not time these three final tasks because we focused on a single function The manual process To execute the test follow these instructions You will need one stopwatch 1 Reboot the system 2 Open Computing pdf 3 Select File Reduce File Size Principled Technologies Inc Total cost of ownership analysis for various computing models 23 Supporting data for performance and sizing calculations In Reduce File Size dialog select Acrobat 7 0 or later Press OK In the Save As dialog give the new file the name Compress Start the Acrobat stopwatch and press Save Stop the Acrobat stopwatch when a Conversion Warning dialog displays Press Enter to close the Conversion Warning dialog 10 Exit Acrobat 11 Delete Compress pdf 12 Empty the Recycle Bin COONS Multitasking scenario Changing the view in a Microsoft PowerPoint presentation while compressing a folder in Windows Explorer The applications involved e Microsoft Windows Explorer for Windows XP Professional Service Pack 2 e Microsoft Office PowerPoint 2003 The data files involved e FourFiles a 265MB folder that the scenario uses to create a 168MB compressed zipped folder on the file server for the thin clients local for the rich clients e Content ppt a 30 4MB Microsoft PowerPoint presentation on the file server The script The script for the scenario performs the
44. ing model This section provides the test execution preparation steps specific to the Terminal Presentation server test bed Getting the Terminal Presentation server network systems ready to go Follow these steps to ready the file server Citrix server and the Terminal Presentation server clients 1 Power on the file server 2 Power on the Citrix server 3 When the servers are fully active power on the Wyse systems you are testing Rebooting the Terminal Presentation server test systems Follow these steps to ready the file server the Citrix server and the Terminal Presentation server clients for testing 1 Click Start gt Restart on one Terminal Presentation server client to reset the Citrix server 2 Once all the Wyse systems are disconnected from the Citrix server shut down the Wyse Clients a Click Start gt Log Off b Choose the Shut Down the System option c Press Enter Log onto the file server as administrator Click Start gt Shut Down Make sure that in the list box under What do you want the computer to do you have selected Restart Type an explanatory comment in the comment box Press Enter PC GO Principled Technologies Inc Total cost of ownership analysis for various computing models 20 Supporting data for performance and sizing calculations 8 Once the servers are fully active power on the Wyse systems you are testing Testing the Typically managed rich desktop clients This section provides the test ex
45. ions for setting up the test beds The Running the tests section gives instructions for running the automated test scripts The final section Application scenarios Descriptions and instructions for running manually details the steps we followed when we hand timed the scenarios to confirm the automated results The automated scripts mimic these steps Appendix A includes test bed diagrams and Appendix B gives detailed descriptions of the clients and servers Principled Technologies Inc Total cost of ownership analysis for various computing models 2 Supporting data for performance and sizing calculations Computing model enterprise view Includes the rich and or thin clients plus any blade PCs or servers the thin 2 D clients connect to in order s S to access the corporate gt gt Q desktop or corporate lications Rich client Wyse thin client SS Client management server s d Access server s se Tet f F ny servers that Figure 2 A model of an enterprise infrastructure including client management infrastructure client access infrastructure and application infrastructure Each test bed represented a subset of an enterprise computing system We assumed the enterprise compute model included three infrastructures e Access infrastructure made up of clients Depending on the computing model these would include thin clients rich clients and blade PCs The access infrastructure also includes for some computing
46. k Change the way Security Center Alerts me on the left d Uncheck Firewall Automatic Updates and Virus Protection 4 Turn off Windows Automatic Updates Doing so prevents such updates from occurring during testing and affecting results a Open the Control Panel b Choose Automatic Updates c Select Turn off Automatic Updates 5 Turn off System Restore Doing so prevents such events from occurring during testing and affecting results a Open the Control Panel b Choose System c Choose the System Restore tab d Select Turn off System Restore on all drives 6 Change the client s IP address to 192 168 1 5 X where X is the client number of the PC We numbered the PCs 1 through 10 with the last address being 192 168 1 60 Click Start gt Control Panel gt Select Network Connections Select Local Area Connection In the Local Area Connection Status dialog select Properties In the Local Area Connection Properties dialog select Properties Select the Use the following IP address radio button Enter the IP Address Select OK on the top two open dialog boxes Select Close to close the third dialog box 7 Map the network drive Q to servername fileserver user files user X where X is as above a Right click My Computer b Select Map Network Drive c Select Q from the Drive drop down menu d Type servername fileserver user files user X where X is as above as the Folder name sa gt o9ao0 D Principled Technologies
47. lains how we manually tested and timed each of those scenarios Though the vast majority of our discussions in this report focus on the results of the automated tests we verified that manually performing the same functions yielded results similar to those of the automated scripts As the instructions below reflect to get the most consistent possible timings and to make our hand timed actions more like the ones the automated scripts perform we sometimes followed procedures for launching applications that were different from those typical users would follow In these cases we also independently verified that the typical user procedures would still show similar results Consequently we are confident that our measurements are accurate and are not artifacts of the measurement or scripting technology Principled Technologies Inc Total cost of ownership analysis for various computing models 22 Supporting data for performance and sizing calculations We ran all application scenarios five times on each of the systems under test and we reported the median of those runs The following subsections which assume you have already completed all of the setup work in the Test methodology section describe how to run each of the individual scenarios by hand Single task scenario Calculating subtotals in Microsoft Excel The application involved e Microsoft Office Excel 2003 The data file involved e Sales2002a1 xls a 1 79MB Excel worksheet located
48. le clients running simultaneously We installed the Microsoft Office 2003 and Adobe Acrobat 8 0 Professional applications so that they would be available to all the clients The test scripts run tasks in these applications Because the thin clients do not have disks all their applications and data files reside on the servers appropriate to the configuration in use We installed the PC applications on each PC but to reflect the nature of today s office environments we stored the data files on the file server except for the Explorer compress and PowerPoint change view test where storing a file locally made more sense in the usage model We ran the three test scripts on each test bed with four client configurations first with a single client running the script then with 2 5 and 10 clients simultaneously running the script This approach allowed us to gauge the response time effects on end users of adding clients to each test bed The rest of this report describes the test environment tests results and analysis The Compute model enterprise view section describes the general test bed configuration The Test results section reports performance and server CPU utilization during the three tests The Results analysis section shows the calculations and estimates for the number of clients per access infrastructure server and for lost user productivity due to server congestion on the access infrastructure servers The Test bed setup section gives instruct
49. lerts me on the left d Uncheck Firewall Automatic Updates and Virus Protection 6 Turn off Windows Automatic Updates Doing so prevents such updates from occurring during testing and affecting results a Open the Control Panel b Choose Automatic Updates c Select Turn off Automatic Updates 7 Turn off System Restore Doing so prevents such events from occurring during testing and affecting results a Open the Control Panel b Choose System c Choose the System Restore tab d Select Turn off System Restore on all drives 8 Change the PC s IP address to 192 168 1 5 X where X is the client number of the PC We numbered the PCs 1 through 10 with the last address being 192 168 1 60 a Click Start gt Control Panel gt Select Network Connections b Select Local Area Connection Principled Technologies Inc Total cost of ownership analysis for various computing models 12 Supporting data for performance and sizing calculations In the Local Area Connection Status dialog select Properties In the Local Area Connection Properties dialog select Properties Select the Use the following IP address radio button Enter the IP Address Select OK on the top two open dialog boxes Select Close to close the third dialog box ap the network drive Q to servername fileserver user files user X where X is as above Right click My Computer Select Map Network Drive Select Q from the Drive drop down menu Type servername f
50. lients for testing 1 Power on the file server 2 Power on the CCI blades that you will be testing 3 When the servers are fully active power on the Wyse systems you are testing Principled Technologies Inc Total cost of ownership analysis for various computing models 21 Supporting data for performance and sizing calculations Rebooting the Blade PC desktop test systems Follow these steps to ready the file server the Citrix server and the Terminal Presentation server clients for testing Double click on the Restart batch file on the desktop of the Wyse systems to reset the CCI blades Click Start gt Log Off and choose Shut Down the System Log onto the file server as administrator Click Start gt Shut Down Make sure that in the list box under What do you want the computer to do you have selected Restart Type an explanatory comment in the comment box Press Enter You may see a warning window that says shutting down will close the remote connections and that asks if you want to end them Click Yes and the file server will restart 9 When the servers are fully active power on the Wyse systems you are testing 00 Or Go N Testing the Well managed application streaming desktop vPro computing model This section provides the test execution preparation steps specific to the Well managed application streaming desktop vPro test bed Getting the Well managed application streaming desktop vPro test bed systems ready to go
51. n Servername Click Start gt Shut Down Make sure that in the list box under What do you want the computer to do you have selected Restart Type an explanatory comment in the comment box Press enter You may see a warning window that says shutting down will close the remote connections and asks if you want to end them Click Yes and the file server will restart Setting up the Typically managed rich desktop test bed We set up the Typically managed rich desktop and test bed by arranging the 10 desktop systems in a network with a 100Mbps switch We then connected this switch through a 1Gbps uplink to the file server Setting up the access infrastructure server in the Typically managed rich desktop test bed The Typically managed rich desktop test bed does not require any servers beyond the file server Setting up the PC clients Repeat these steps on each of the 10 PC clients 2 Install Windows XP Professional on each of the clients Apply all XP updates as of July 10 2007 from the Microsoft Windows Update Web site including Windows XP SP2 3 Install Microsoft Office 2003 4 Apply all Office 2003 updates as of July 10 2007 from the Microsoft Office Update Web site 5 Turn off Windows Service Pack 2 Security Center Pop up Alerts Doing so prevents such alerts from occurring during testing and affecting results a Open the Control Panel b Choose Security Center c Click Change the way Security Center A
52. nd benefits of typically managed compared to well managed models and the costs and benefits of Intel vPro technology We refer to these features in model names for consistency with other related reports even though they do not affect the platform performance and sizing that is the focus of this document In this context an access infrastructure server is one that does the work that would be done on a rich client in a traditional environment Server congestion occurs when the combined needs of the many clients accessing one server exceed the available server resources Server congestion is not applicable to the two rich desktop or the Blade PC desktop computing models which have a one to one ratio of users to devices Each of our test beds included a file server and 10 client systems We used a pair of identical file servers to allow us to test two computing models at a time Our hand built file server included a 3 4 GHz Intel Xeon processor 2GB DDR2 SRAM an HP 382146 405 motherboard and a Western Digital WD740ADFD OONLR1 disk and ran Windows Server 2003 R2 Enterprise with Service Pack 2 We used a 100Mbps network infrastructure for the clients and a 1Gbps network infrastructure for the servers to reflect typical scenarios in enterprises today We also set up access infrastructure servers for computing models that require them Our hand built access infrastructure server included two 3GHz Intel Xeon 5160 dual core processors 16GB DDR2 SRAM a Supermicro
53. providing the single client s level of performance 3 Measuring network performance during a typical test We used our Adobe Acrobat test and a formula very similar to that in case 2 above number of real number of clients x clients X OS streaming single client test duration percentage network each test client OS streaming 10 client test duration utilization represents raw client count client factor OS streaming 10 client test time normalized to the OS streaming single estimated number of real clients client test time Raw client count To determine raw client count we divided the number of clients simultaneously executing network intensive tasks by the peak percentage network utilization during the test For our analysis we used the 10 client case so the number of clients was always 10 Client factor All of our test clients simultaneously performed network intensive tasks To account for the fact that only a percentage of any real life group of clients would be executing these tasks at any given time we assigned a Client factor the number of real clients each client in our tests represented Our default client factor was 10 reflecting an industry standard range of 8 to 10 Estimated number of real clients The product of the raw client count and the client factor was the estimated number of real clients the computing model could support if all computing models executed the tasks in the same amount of time Because this was not
54. results a Open the Control Panel b Choose Automatic Updates c Select Turn off Automatic Updates 5 Turn off System Restore Doing so prevents such events from occurring during testing and affecting results a Open the Control Panel b Choose System c Choose the System Restore tab d Select Turn off System Restore on all drives Start the Windows DHCP service for the server Install Ardence server making sure that every option is set so that the local server handles it Start Ardence Administrator Determine which computer will act as the prime client and manually add its MAC address to Ardence Administrator s client list 10 To ensure as consistent a starting point as possible for the performance measurements defragment the hard disk 11 Using Symantec s Ghost utility make an image of the hard disk This image let us return to a clean and consistent starting point whenever necessary OOND Setting up the Well managed OS streaming desktop vPro clients Repeat these steps on each of the 10 Ardence clients 1 Change the system s boot order so that network boot will be the first method listed a Turnon the system and access the BIOS using the appropriate key b Using the directional arrows on the keyboard highlight the boot order option c Following the directions on the screen place network boot in the number one position d Save your changes and exit the BIOS Additional client setup The following setup is to
55. rich client HP BladeSystem bc2000 Blade PC Wyse Winterm V50 thin client Memory module s Vendor and model number Kingston KHX6400D2K2 1G Hyundai Electronics HYMP564S64CP6 Y5 DSL CL2 5 Type Speed MHz Speed running in the system MHz Timing Latency tCL tRCD tRP tRASmin PC2 5300 DDR2 667 333 5 5 5 15 PC2 5300 DDR2 667 333 5 5 5 15 PC 2700 DDR 667 333 Size 1 024 MB 1 024 MB 256 MB Number of memory modules 2x 512 MB 2x 512 MB 1 Chip organization Single Sided Double sided Double sided Channel Dual Dual Single Hard disk or flash memory Vendor and model number WD Caviar WD1600JS 22MHB1 Samsung HMO80HI Samsung Flash Memory P770478 10LQ Size 160 GB 80 GB 128 MB Buffer size 8 MB 8 MB N A RPM 7 200 7 200 N A Type SATA SATA N A Controller Intel 82801HO ICH8D0 ATI TechIXP SB600 N A Controller driver Microsoft 5 1 2535 0 Microsoft 5 1 2535 0 N A Operating system Name Build number Microsoft Windows XP Professional 5 01 2600 Microsoft Windows XP Professional 5 01 2600 Wyse Linux V6 6 3 1 Service pack SP2 SP2 File system NTFS NTFS Kernel ACPI Multiprocessor PC ACPI Uniprocessor PC 2 6x Language English English English Microsoft DirectX version
56. roductivity per day and the number of clients an access infrastructure server can support The Results analysis section describes those calculations Number of clients Terminal Presentation server Typically managed rich desktop Well managed OS streaming desktop vPro Blade PC desktop Well managed application streaming desktop vPro Figure 3 Sum of script execution times for all tests all times in seconds lower is better Results analysis Our aim was to estimate for each computing model the lost productivity per user per day and the number of clients an access infrastructure server can realistically support We relied on the test results in the section above in making these estimates The workers in the example enterprise are knowledge workers Knowledge workers perform a mixture of high medium and low demand tasks primarily medium demand tasks Their demand is higher than light or transactional users who use few applications to carry out repetitive low demand tasks and lower than power or heavy users who typically run high demand applications such as 3 D design applications and complex financial applications Knowledge workers typically use a variety of applications in the case of this enterprise a total of 120 applications among all the users They also frequently leave multiple applications open simultaneously and Principled Technologies Inc Total cost of ownership analysis for various computin
57. rver 2003 Update KB932168 Windows Server 2003 Security Update KB932168 Windows Server 2003 Security Update KB933566 Windows Server 2003 Security Update KB935839 Windows Server 2003 Security Update KB935840 Windows Server 2003 Security Update KB924667 v2 Windows Server 2003 Security Update KB925902 3 Turn off Windows Service Pack 2 Security Center Pop up Alerts Doing so prevents such alerts from occurring during testing and affecting results a Open the Control Panel b Choose Security Center c Click Change the way Security Center Alerts me on the left d Uncheck Firewall Automatic Updates and Virus Protection 4 Install Microsoft Office 2003 5 Apply all Office 2003 updates as of July 10 2007 from the Microsoft Office Update Web site Principled Technologies Inc Total cost of ownership analysis for various computing models 10 Supporting data for performance and sizing calculations 6 Turn off Windows Automatic Updates Doing so prevents such updates from occurring during testing and affecting results a Open the Control Panel b Choose Automatic Updates c Select Turn off Automatic Updates 7 Turn off System Restore Doing so prevents such events from occurring during testing and affecting results a Open the Control Panel b Choose System c Choose the System Restore tab d Select Turn off System Restore on all drives 8 Create 10 users USER1 through USER10 Give each user remote desktop privileges an
58. ser files user X where X is as above a Right click My Computer b Select Map Network Drive c Select Q from the Drive drop down menu d Type servername fileserver user files user X where X is as above as the Folder name e Click Finish 9 Create a desktop shortcut to the Q network drive This shortcut s name will be of the form Shortcut to UserX on servername 10 Configure the Visual Test Runtime application so that it will work with all the test scripts a Copy the following five Visual Test dll files into WINDOWS SYSTEM32 e EHelper dll e 6Vtaa dll e VTest60 dll e 6Vtres dll e WebDrive dll b Open a command prompt c Type cd WINDOWS SYSTEM32 and press Enter d For each of the following three dlls type regsvr32 dll filename and press Enter This command registers a dll with the system e EHelper dll e Vtaa dll e WebDrive dll 11 Install Adobe Acrobat 8 0 Professional 12 Install Ardence Client on the system 13 To ensure as consistent a starting point as possible for the performance measurements defragment the hard disk of each PC 14 Restart the client making sure that it restarts locally a In Ardence Server Administrator set the client that contains the Windows Office Acrobat applications to force a local boot 15 After rebooting go through the Start menu and start the image creation process 16 After creating the vDisk image log onto the server and start up Ar
59. servers directory servers Web servers middleware servers print servers file servers etc For our tests we use a file server as x representative While File server s all client types strain the application infrastructure the load does not differ significantly by client type Figure 9 Typically managed rich desktop test bed Principled Technologies Inc Total cost of ownership analysis for various computing models 27 Supporting data for performance and sizing calculations Well managed OS streaming desktop vPro test bed Figure 10 illustrates the Well managed OS streaming desktop vPro test bed which includes the following components e Rich Clients hand built Processor 2 13 GHz Intel Core 2 Duo E6400 Memory 1GB DDR2 SRAM Motherboard Intel DQ965GF Disk 160 GB WD Caviar WD1600JS Operating System Windows XP Professional version 2002 with Service Pack 2 Streaming software Ardence Evaluation Client 4 1 000000 e Access infrastructure server hand built o Processors Two 3GHz Intel Xeon 5160 dual core processors Memory 16GB DDR2 SRAM Motherboard Supermicro X7DBE Disks Four 73GB Seagate Cheetah ST373455SS disks Operating system Windows Server 2003 R2 Enterprise with Service Pack 2 Streaming software Ardence Evaluation Server 4 1 OS 00000 e File Server hand built o Processors One 3 4GHz Intel Xeon processor o Memory 2GB DDR2 SRAM o Motherboard HP 382146 405 o Disks One Western D
60. servers are fully active power on the PCs you are testing Rebooting the Well managed OS streaming desktop vPro test bed systems Follow these steps to ready the file server Ardence server and Ardence PCs you are testing 1 Click Start gt Shut Down on each active PC Select Shut Down from the menu on each active PC Log on to the file server as administrator Click Start gt Shut Down Make sure that in the list box under What do you want the computer to do you have selected Restart Type an explanatory comment in the comment box Press Enter You may see a warning window that says shutting down will close the remote connections and that asks if you want to end them Click Yes and the file server will restart 9 Log onto the Ardence server as administrator 10 Click Start gt Shut Down 11 Make sure that in the list box under What do you want the computer to do you have selected Restart 12 Press Enter 13 You may see a warning window that says shutting down will close the remote connections and that asks if you want to end them Click Yes and the server will restart 14 When the servers are fully active power on the PCs you are testing ONAARWN Testing the Blade PC desktop computing model This section provides the test execution preparation steps specific to the Blade PC desktop test bed Getting the Blade PC desktop test bed systems ready to go Follow these steps to ready the files server CCI server and CCI c
61. t count To determine raw client count we divided the number of clients simultaneously executing network intensive tasks by the peak percentage network utilization during the boot cycle For our analysis we used the 10 client case so the number of clients was always 10 Client factor All of our test clients simultaneously performed network intensive tasks To account for the fact that only a percentage of any real life group of clients would be executing these tasks at any given time we assigned a client factor the number of actual clients each client in our tests represented Our default client factor was 10 reflecting an industry standard range of 8 to 10 Estimated number of real clients The product of the raw client count and the client factor was the estimated number of real clients the computing model could support if all computing models executed the tasks in the same amount of time Because this was not the case we used one additional factor to account for the variation in speed Principled Technologies Inc Total cost of ownership analysis for various computing models 7 Supporting data for performance and sizing calculations OS streaming 10 client boot time normalized to the OS streaming single client boot time For the last factor we divided the boot completion time of a single client by the 10 client boot completion time Normalizing to the single client performance estimated the number of clients the server could handle while
62. tes the Blade PC desktop test bed which includes the following components e Blade PCs HP BladeSystem bc2000 Blade PC o Processor AMD Athlon 64 2100 1 20GHz o Memory 1GB DDR2 RAM o Disk 80 GB Samsung HMO80HI o Operating system Windows XP Professional version 2002 with Service Pack 2 e Thin Clients Wyse Winterm V50 o Processor 1GHz Via C3 Memory 128MB flash 256 MB DDR RAM o Operating system Wyse Linux V6 e File Server hand built o Processors One 3 4GHz Intel Xeon processor Memory 2GB DDR2 SRAM Motherboard HP 382146 405 Disks One Western Digital WD740ADFD 00NLR1 Operating system Windows Server 2003 R2 Enterprise with Service Pack 2 0000 Note We assigned each thin client a blade PC via a static IP address No session manager was running Blade PCs require an access infrastructure with a Includes the rich and or thin blade PC for every clients plus any blade PCs or active thin client 2 servers the thin clients session to access gt S connect to in order to the corporate access the corporate infrastructure Wyse thin client desktop or corporate 1 through 10 applications From an access All computations infrastructure point and performance of view concurrent measurements take 5 CCI clients scale place directly on the HP CCI enclosure directly with the individual blades with bc2000 blade number of blade PCs available Includes corporate servers that support both thin and rich clients such as database
63. th Service Pack 2 All thin clients were Wyse Winterm V50 thin clients For the Blade PC desktop model we used HP BladeSystem bc2000 Blade PCs AMD Athlon 64 2100 1 20GHz 1GB DDR2 RAM running Windows XP Professional version 2002 with Service Pack 2 We collected performance data and server CPU utilization data while running compute intensive tasks on each computing model For models that require access infrastructure servers we used the data to estimate the number of clients that a server can realistically support Rich desktop models showed the fastest performance computing models that use access infrastructure servers for compute tasks were slower at higher client counts due to server congestion the Blade PC platform was relatively slow at all client counts We used the performance data and estimates of the frequency of the compute intensive tasks that would cause model slowness to estimate minutes of lost productivity per user per day due to platform slowness Figure 1 shows these estimates Terminal Presentation server Typically managed rich desktop Well managed OS streaming desktop vPro Blade PC desktop Well managed Application streaming desktop vPro Figure 1 Estimated number of clients per access infrastructure server and minutes of lost productivity per user per day due to platform slowness for various computing models i These tests were part of a larger computing model comparison that considered the costs a
64. ulations Figure 5 shows the results of these calculations Note For Well managed application streaming desktop vPro we rounded the shared server minutes lost per day results to zero Terminal Presentation server Typically managed rich desktop Well managed OS streaming desktop vPro Blade PC desktop Well managed Application streaming desktop vPro Figure 5 Lost user productivity lost per day in minutes Number of clients an access infrastructure server can support To calculate the number of clients an access infrastructure server can support in comparison to rich client performance we used the following formula number of actual number of clients x clients x rich client time percentage CPU utilization each test client shared platform time represents raw client count client factor shared platform speed estimated number of real clients normalized to rich client The italicized labels explain what each section of the formula represents Raw client count To determine raw client count we divided the number of clients simultaneously executing server intensive tasks by the average percentage of CPU utilization during the test For our analysis we used the 10 client case so the number of clients was always 10 Figure 6 shows average access infrastructure server CPU utilization for each computing model during a 10 client run of each of the three tests Terminal Presentation server Typically
65. x B Detailed test bed configurations Servers Servers Supermicro X7DB8 access infrastructure server HP ProLiant DL360 file server General Processor and OS kernel physical core logical UP MP 2P 4C 4L MP 1P 1C 2L MP Number of physical processors 2 1 Single dual core processors Dual Single System Power Management Policy Always on Always on CPU Vendor Intel Intel Name Intel Xeon 5160 Intel Xeon Stepping 4 A Socket type 771 LGA 604 mPGA Core frequency GHz 3 0 3 4 Front side bus frequency MHz 1 333 800 L1 cache 32 KB 32 KB per core 16 KB 12 Kuops L2 cache 4 096 KB shared 2 048 KB Platform Vendor and model number Supermicro HP Motherboard model number Supermicro X7DB8 382146 405 Motherboard chipset Intel 5000P Intel E7520 Motherboard revision number 92 A05 Motherboard serial number TM66S06522 USE617N2DF BIOS name and version Phoenix Technologies LTD 6 00 HP P54 BIOS settings Default Default Memory module s Vendor and model number Kingston KVR667D2D8F5 1G Infineon HYS72T28000HR 5 A Type PC2 5300 FB DDR2 PC2 3200 DDR2 Speed MHz 667 400 Speed in the system currently running MHz 333 200 Timing latency tCL tRCD tRP tRASmin 5 5

Total cost of ownership analysis for various computing models

Contents

Download Pdf Manuals

Related Search

Related Contents