SlideShare a Scribd company logo

Wp intelli cache_reduction_iops_xd5.6_fp1_xs6.1

•
•
Nuno Alves
Nuno Alves

CTX138217 - IntelliCache Reduction in IOPS: XenDesktop 5.6 FP1 on XenServer 6.1 - Citrix Knowledge Center http://ow.ly/o3Ma4 The purpose of this document is to provide testing results based on MCS-delivered streamed virtual desktops leveraging IntelliCache

1 of 12
Download to read offline
1 IntelliCache Reduction in IOPS: XenDesktop 5.6 FP1 on XenServer 6.1
2 Table of Contents Objective.......................................................................................................................................................3 Hardware ......................................................................................................................................................3 Common Testing Infrastructure....................................................................................................................4 Virtual Desktop Software Configuration.......................................................................................................4 Virtual Desktop Hardware Configuration .....................................................................................................4 Test Methodology.........................................................................................................................................4 IntelliCache Stages........................................................................................................................................5 Write Cache Non-Persist:..........................................................................................................................5 1. Cold Cache: ...................................................................................................................................5 2. Warm Cache:.................................................................................................................................5 Write Cache Persist:..................................................................................................................................5 Results...........................................................................................................................................................5 Results from Half Enclosure (8 Node, Single Pool) Testing:......................................................................6 IOPS: Central Storage – Non-IntelliCache Baseline...............................................................................6 IOPS: Central Storage – IntelliCache Persist .........................................................................................6 Results from Full Enclosure (16 Node, Dual Pool) Testing:.......................................................................7 IOPS: Central Storage – Non-IntelliCache Baseline...............................................................................7 IOPS: Central Storage – IntelliCache Persist .........................................................................................7 Conclusion:....................................................................................................................................................8 Supplemental:...............................................................................................................................................9 Login VSI: Results from Half Enclosure Non-IntelliCache Baseline (8 Node, Single Pool) Testing:...........9 Login VSI: Results from Half Enclosure IntelliCache Persist (8 Node, Single Pool) Testing:....................10 Login VSI: Results from Full Enclosure Non-IntelliCache Baseline (16 Node, Dual Pool) Testing:..........11 Login VSI: Results from Full Enclosure IntelliCache Persist (16 Node, Dual Pool) Testing: ....................12
3 Objective The purpose of this document is to provide testing results based on MCS-delivered streamed virtual desktops leveraging IntelliCache. The data within this document is based on a half enclosure test of 8 nodes in a single pool, a SMB (Small Business) deployment, and a full enclosure test of 16 nodes divided into two pools, a SSE (Small-Scale Enterprise) deployment. The components that were used in both of these environments included XenServer 6.1 (for the virtualization infrastructure) including up to hotfix E003, XenDesktop 5.6 FP1 Platinum Edition, Licensing Server 11.10 VPX, Web Interface 5.4, UPM 4.1.1.5 and SQL Server 2008 R2. Hardware The information below outlines the type of hardware that was used for the specific area of the environment and what server roles used which hardware: Blade Enclosure Model# Blade System c7000 Platinum Enclosure G2 Firmware 3.70 Oct 01 2012 XenServer Hosts Model# ProLiant BL460c Gen 8 Blade Server CPU Dual Eight (8) Core CPU @ 2.60Ghz Intel Xeon E5-2670 Memory 256Gb Storage Two (2) 400GB SSD (RAID 0) Network HP FlexFabric 10Gb 2-port 554FLB Adapter Firmware I31 02/25/2012 AD, STAT, PVS and XS (Infrastructure) Model# ProLiant BL460c Gen 7 Blade Server CPU Dual Six (6) Core CPU @ 2.67Ghz Intel Xeon X5650 Memory 192Gb Storage Two (2) 136GB (RAID 1) Network FlexFabric Embedded Ethernet Firmware I27 05/05/2011 VDA Storage Type Storage Area Network (SAN) Raid Configuration Raid DP (Raid-6) 64-bit – 20 Disk Aggregate (2 Raid Groups) Disk Drives 3.5” SAS 15K RPM 450GB Networking Dual 10GbE
4 Common Testing Infrastructure The common infrastructure is defined as the group of core component servers that were directly related to both of the testing scenarios. This infrastructure was made up of the following components:  2 Active Directory controllers: One (1) physical server and one (1) virtual machine  1 STAT, SQL & VSI file share server: One (1) physical server  1 PVS server: One (1) physical server (hosting client launchers and XenMetrics virtual machines)  1 PVS SQL Database server: One (1) virtual machine  2 servers running XenServer 6.02 (for STAT client and XenMetrics hosting)  1 XenDesktop 5.6 FP1 DDC server: One (1) virtual machine  1 XenDesktop SQL Database server: One (1) virtual machine  1 License Server: One (1) virtual machine  2 Central Storage PerfMon Collectors: Two (2) virtual machines NOTE: This infrastructure was shared and unchanged per test phase. Virtual Desktop Software Configuration  Windows 7 Enterprise SP1 x86 (32-bit)  Office 2010 Professional SP1 x86 (32-bit)  LoginVSI 3.6  Citrix Profile Management Agent 4.1.1.5  Citrix Virtual Desktop Agent 5.6 Virtual Desktop Hardware Configuration VDA Configuration OS Windows 7 Enterprise SP1 x86 HDD 40GB HDD #2 1GB (Differential) HDD #3 16MB (Identity) vCPU 1 Memory 1024MB SWAP 2048MB Test Methodology The goal of these tests were to determine scalability on the HP BL460c G8 blades as defined above. Our first test included a half enclosure (8 nodes deployed in a single pool) with 1,350 VDAs equally deployed. From there we moved on to a full enclosure (16 nodes divided into two pools) test with 2,700 VDAs equally deployed. Tests were conducted using a third party tool called LoginVSI 3.6. This application runs a predefined VSI medium workload. The session’s response times were monitored via this tool to produce a score, VSI Max score, designed to give the maximum number of desktops a given host could support. The total number of sessions (VSI Max) would equate to the maximum VM density in a single server scenario. STAT, an internal performance tool, was used as a client launching mechanism to drive these sessions, capture system performance and monitor their state.
5 IntelliCache Stages Write Cache Non-Persist: 1. Cold Cache: For the first Login VSI test run, we used VMs that were booted and left to idle for 1 hour. After idling, Login VSI launched users to begin testing. Once the users had logged in, the Login VSI medium workload is started. After all users have successfully run the workload, each user is then automatically logged off. XenServer stores the data read from the shared storage in the local Read Cache. 2. Warm Cache: After XenServer populates the Read Cache with the initial log‐on data, the VMs can then obtain most of their data from the local Read Cache, but it is still not completely populated. After idling for another hour, the second Login VSI test is now executed. Continuing with a medium workload, XenServer stores any remaining data read from the shared storage in the local Read Cache. The local Read Cache is now populated and ready for fully persistent testing. Write Cache Persist: 3. After the VMs are idle for another hour, the final Login VSI medium workload test is now executed, fully utilizing the populated XenServer Read Cache. This is the test that we utilize for our IntelliCache performance figures. Results In our half enclosure environment, enabling IntelliCache resulted in a 99.9% decrease in IOPS on shared storage with a decrease from 21,799 IOPS to just 21 IOPS. In our full enclosure environment, enabling IntelliCache resulted in a 99.7% decrease in IOPS on shared storage with a decrease from 39,207 IOPS to just 126 IOPS. To understand these findings, we will outline them on the next page.
6 Results from Half Enclosure (8 Node, Single Pool) Testing: IOPS: Central Storage – Non-IntelliCache Baseline IOPS: Central Storage – IntelliCache Persist *This data was collected directly from the central storage device during the VSI Medium Workload Test. 22728.98 10819.99 10979.38 0.00 5000.00 10000.00 15000.00 20000.00 25000.00 Baseline Half Enclosure (1,350 VDA) Baseline Test Max Total IOPS Max Read IOPS Max Write IOPS 51.13 22.18 15.31 11.72 1.29 5.86 10.58 22.18 15.31 0.00 10.00 20.00 30.00 40.00 50.00 60.00 IntelliCache Cold Cache IntelliCache Warm Cache IntelliCache Persist Half Enclosure (1,350 VDA) IntelliCache Test Max Total IOPS Max Read IOPS Max Write IOPS
7 Results from Full Enclosure (16 Node, Dual Pool) Testing: IOPS: Central Storage – Non-IntelliCache Baseline NOTE: The central storage device was at max CPU load during our dual pool baseline (non-IntelliCache) testing. IOPS: Central Storage – IntelliCache Persist NOTE: The central storage device was at max CPU load during our dual pool (IntelliCache Cold Cache) testing. *This data was collected directly from the central storage device during the VSI Medium Workload Test. 29281.26 23455.11 15751.61 0.00 5000.00 10000.00 15000.00 20000.00 25000.00 30000.00 35000.00 Baseline XS 6.1 Dual Pool (2,700 VDA) Non-IntelliCache Baseline Testing Max Total IOPS Max Read IOPS Max Write IOPS 3188.63 103.21 93.63 3183.35 8.57 88.6427.57 103.21 37.73 0.00 500.00 1000.00 1500.00 2000.00 2500.00 3000.00 3500.00 IntelliCache Cold Cache IntelliCache Warm Cache IntelliCache Persist XS 6.1 Dual Pool (2,700 VDA) IntelliCache Testing Max Total IOPS Max Read IOPS Max Write IOPS
8 Conclusion: Based on our half enclosure (1,350 VDA) and full enclosure (2,700 VDA) tests, we can conclude that by enabling IntelliCache we see about a 100% reduction in IOPS on the shared storage system after the read cache is fully populated on the local XenServer host. Scaling out the XenServer host beyond Single Server as previously documented by Jeffry Kuhn* shows absolutely no loss in reduction of IOPS when utilized in SMB (Small Business) and SSE (Small-Scale Enterprise) deployments. *PLEASE REFER TO CTX134613 - OPTIMIZING STORAGE FOR XENDESKTOP WITH XENSERVER INTELLICACHE
9 Supplemental: Login VSI: Results from Half Enclosure Non-IntelliCache Baseline (8 Node, Single Pool) Testing: VSI Detailed: VSI Scatter:
10 Login VSI: Results from Half Enclosure IntelliCache Persist (8 Node, Single Pool) Testing: VSI Detailed: VSI Scatter:
11 Login VSI: Results from Full Enclosure Non-IntelliCache Baseline (16 Node, Dual Pool) Testing: VSI Detailed: VSI Scatter:
12 Login VSI: Results from Full Enclosure IntelliCache Persist (16 Node, Dual Pool) Testing: VSI Detailed: VSI Scatter:

More Related Content

Wp intelli cache_reduction_iops_xd5.6_fp1_xs6.1

  • 1. 1 IntelliCache Reduction in IOPS: XenDesktop 5.6 FP1 on XenServer 6.1
  • 2. 2 Table of Contents Objective.......................................................................................................................................................3 Hardware ......................................................................................................................................................3 Common Testing Infrastructure....................................................................................................................4 Virtual Desktop Software Configuration.......................................................................................................4 Virtual Desktop Hardware Configuration .....................................................................................................4 Test Methodology.........................................................................................................................................4 IntelliCache Stages........................................................................................................................................5 Write Cache Non-Persist:..........................................................................................................................5 1. Cold Cache: ...................................................................................................................................5 2. Warm Cache:.................................................................................................................................5 Write Cache Persist:..................................................................................................................................5 Results...........................................................................................................................................................5 Results from Half Enclosure (8 Node, Single Pool) Testing:......................................................................6 IOPS: Central Storage – Non-IntelliCache Baseline...............................................................................6 IOPS: Central Storage – IntelliCache Persist .........................................................................................6 Results from Full Enclosure (16 Node, Dual Pool) Testing:.......................................................................7 IOPS: Central Storage – Non-IntelliCache Baseline...............................................................................7 IOPS: Central Storage – IntelliCache Persist .........................................................................................7 Conclusion:....................................................................................................................................................8 Supplemental:...............................................................................................................................................9 Login VSI: Results from Half Enclosure Non-IntelliCache Baseline (8 Node, Single Pool) Testing:...........9 Login VSI: Results from Half Enclosure IntelliCache Persist (8 Node, Single Pool) Testing:....................10 Login VSI: Results from Full Enclosure Non-IntelliCache Baseline (16 Node, Dual Pool) Testing:..........11 Login VSI: Results from Full Enclosure IntelliCache Persist (16 Node, Dual Pool) Testing: ....................12
  • 3. 3 Objective The purpose of this document is to provide testing results based on MCS-delivered streamed virtual desktops leveraging IntelliCache. The data within this document is based on a half enclosure test of 8 nodes in a single pool, a SMB (Small Business) deployment, and a full enclosure test of 16 nodes divided into two pools, a SSE (Small-Scale Enterprise) deployment. The components that were used in both of these environments included XenServer 6.1 (for the virtualization infrastructure) including up to hotfix E003, XenDesktop 5.6 FP1 Platinum Edition, Licensing Server 11.10 VPX, Web Interface 5.4, UPM 4.1.1.5 and SQL Server 2008 R2. Hardware The information below outlines the type of hardware that was used for the specific area of the environment and what server roles used which hardware: Blade Enclosure Model# Blade System c7000 Platinum Enclosure G2 Firmware 3.70 Oct 01 2012 XenServer Hosts Model# ProLiant BL460c Gen 8 Blade Server CPU Dual Eight (8) Core CPU @ 2.60Ghz Intel Xeon E5-2670 Memory 256Gb Storage Two (2) 400GB SSD (RAID 0) Network HP FlexFabric 10Gb 2-port 554FLB Adapter Firmware I31 02/25/2012 AD, STAT, PVS and XS (Infrastructure) Model# ProLiant BL460c Gen 7 Blade Server CPU Dual Six (6) Core CPU @ 2.67Ghz Intel Xeon X5650 Memory 192Gb Storage Two (2) 136GB (RAID 1) Network FlexFabric Embedded Ethernet Firmware I27 05/05/2011 VDA Storage Type Storage Area Network (SAN) Raid Configuration Raid DP (Raid-6) 64-bit – 20 Disk Aggregate (2 Raid Groups) Disk Drives 3.5” SAS 15K RPM 450GB Networking Dual 10GbE
  • 4. 4 Common Testing Infrastructure The common infrastructure is defined as the group of core component servers that were directly related to both of the testing scenarios. This infrastructure was made up of the following components:  2 Active Directory controllers: One (1) physical server and one (1) virtual machine  1 STAT, SQL & VSI file share server: One (1) physical server  1 PVS server: One (1) physical server (hosting client launchers and XenMetrics virtual machines)  1 PVS SQL Database server: One (1) virtual machine  2 servers running XenServer 6.02 (for STAT client and XenMetrics hosting)  1 XenDesktop 5.6 FP1 DDC server: One (1) virtual machine  1 XenDesktop SQL Database server: One (1) virtual machine  1 License Server: One (1) virtual machine  2 Central Storage PerfMon Collectors: Two (2) virtual machines NOTE: This infrastructure was shared and unchanged per test phase. Virtual Desktop Software Configuration  Windows 7 Enterprise SP1 x86 (32-bit)  Office 2010 Professional SP1 x86 (32-bit)  LoginVSI 3.6  Citrix Profile Management Agent 4.1.1.5  Citrix Virtual Desktop Agent 5.6 Virtual Desktop Hardware Configuration VDA Configuration OS Windows 7 Enterprise SP1 x86 HDD 40GB HDD #2 1GB (Differential) HDD #3 16MB (Identity) vCPU 1 Memory 1024MB SWAP 2048MB Test Methodology The goal of these tests were to determine scalability on the HP BL460c G8 blades as defined above. Our first test included a half enclosure (8 nodes deployed in a single pool) with 1,350 VDAs equally deployed. From there we moved on to a full enclosure (16 nodes divided into two pools) test with 2,700 VDAs equally deployed. Tests were conducted using a third party tool called LoginVSI 3.6. This application runs a predefined VSI medium workload. The session’s response times were monitored via this tool to produce a score, VSI Max score, designed to give the maximum number of desktops a given host could support. The total number of sessions (VSI Max) would equate to the maximum VM density in a single server scenario. STAT, an internal performance tool, was used as a client launching mechanism to drive these sessions, capture system performance and monitor their state.
  • 5. 5 IntelliCache Stages Write Cache Non-Persist: 1. Cold Cache: For the first Login VSI test run, we used VMs that were booted and left to idle for 1 hour. After idling, Login VSI launched users to begin testing. Once the users had logged in, the Login VSI medium workload is started. After all users have successfully run the workload, each user is then automatically logged off. XenServer stores the data read from the shared storage in the local Read Cache. 2. Warm Cache: After XenServer populates the Read Cache with the initial log‐on data, the VMs can then obtain most of their data from the local Read Cache, but it is still not completely populated. After idling for another hour, the second Login VSI test is now executed. Continuing with a medium workload, XenServer stores any remaining data read from the shared storage in the local Read Cache. The local Read Cache is now populated and ready for fully persistent testing. Write Cache Persist: 3. After the VMs are idle for another hour, the final Login VSI medium workload test is now executed, fully utilizing the populated XenServer Read Cache. This is the test that we utilize for our IntelliCache performance figures. Results In our half enclosure environment, enabling IntelliCache resulted in a 99.9% decrease in IOPS on shared storage with a decrease from 21,799 IOPS to just 21 IOPS. In our full enclosure environment, enabling IntelliCache resulted in a 99.7% decrease in IOPS on shared storage with a decrease from 39,207 IOPS to just 126 IOPS. To understand these findings, we will outline them on the next page.
  • 6. 6 Results from Half Enclosure (8 Node, Single Pool) Testing: IOPS: Central Storage – Non-IntelliCache Baseline IOPS: Central Storage – IntelliCache Persist *This data was collected directly from the central storage device during the VSI Medium Workload Test. 22728.98 10819.99 10979.38 0.00 5000.00 10000.00 15000.00 20000.00 25000.00 Baseline Half Enclosure (1,350 VDA) Baseline Test Max Total IOPS Max Read IOPS Max Write IOPS 51.13 22.18 15.31 11.72 1.29 5.86 10.58 22.18 15.31 0.00 10.00 20.00 30.00 40.00 50.00 60.00 IntelliCache Cold Cache IntelliCache Warm Cache IntelliCache Persist Half Enclosure (1,350 VDA) IntelliCache Test Max Total IOPS Max Read IOPS Max Write IOPS
  • 7. 7 Results from Full Enclosure (16 Node, Dual Pool) Testing: IOPS: Central Storage – Non-IntelliCache Baseline NOTE: The central storage device was at max CPU load during our dual pool baseline (non-IntelliCache) testing. IOPS: Central Storage – IntelliCache Persist NOTE: The central storage device was at max CPU load during our dual pool (IntelliCache Cold Cache) testing. *This data was collected directly from the central storage device during the VSI Medium Workload Test. 29281.26 23455.11 15751.61 0.00 5000.00 10000.00 15000.00 20000.00 25000.00 30000.00 35000.00 Baseline XS 6.1 Dual Pool (2,700 VDA) Non-IntelliCache Baseline Testing Max Total IOPS Max Read IOPS Max Write IOPS 3188.63 103.21 93.63 3183.35 8.57 88.6427.57 103.21 37.73 0.00 500.00 1000.00 1500.00 2000.00 2500.00 3000.00 3500.00 IntelliCache Cold Cache IntelliCache Warm Cache IntelliCache Persist XS 6.1 Dual Pool (2,700 VDA) IntelliCache Testing Max Total IOPS Max Read IOPS Max Write IOPS
  • 8. 8 Conclusion: Based on our half enclosure (1,350 VDA) and full enclosure (2,700 VDA) tests, we can conclude that by enabling IntelliCache we see about a 100% reduction in IOPS on the shared storage system after the read cache is fully populated on the local XenServer host. Scaling out the XenServer host beyond Single Server as previously documented by Jeffry Kuhn* shows absolutely no loss in reduction of IOPS when utilized in SMB (Small Business) and SSE (Small-Scale Enterprise) deployments. *PLEASE REFER TO CTX134613 - OPTIMIZING STORAGE FOR XENDESKTOP WITH XENSERVER INTELLICACHE
  • 9. 9 Supplemental: Login VSI: Results from Half Enclosure Non-IntelliCache Baseline (8 Node, Single Pool) Testing: VSI Detailed: VSI Scatter:
  • 10. 10 Login VSI: Results from Half Enclosure IntelliCache Persist (8 Node, Single Pool) Testing: VSI Detailed: VSI Scatter:
  • 11. 11 Login VSI: Results from Full Enclosure Non-IntelliCache Baseline (16 Node, Dual Pool) Testing: VSI Detailed: VSI Scatter:
  • 12. 12 Login VSI: Results from Full Enclosure IntelliCache Persist (16 Node, Dual Pool) Testing: VSI Detailed: VSI Scatter: