Transcription

High Performance 750-Seat VirtualDesktop Infrastructure (VDI)with Citrix XenDesktop 7Reference ArchitectureFebruary, 2014

Table of ContentsExecutive Summary . 3Infrastructural Components . 4Citrix XenDesktop . 4Microsoft Windows Server 2012 R2 Hyper-V . 4The X-IO Intelligent Storage Element . 5Login VSI . 6Reference Architecture Configuration . 7Physical Hardware Configuration . 7Virtual Machine Configuration . 8Citrix XenDesktop Storage Requirements . 9Windows Server 2012 R2 Configuration and Best Practices . 9Network Configuration . 10Intelligent Storage Element Test Configuration . 11Fibre Channel Switch Configuration . 11LUN Configuration . 12Login VSI Configuration . 13Test Results . 14ISE 230 Test Results . 15Login VSI Test Results . 18Total Cost of Ownership (TCO) Analysis . 20TCO versus Cost of Acquisition . 20Sizing Storage for Actual Performance Requirements . 20Storage Density and Rack Space Consumption . 21Storage Energy Consumption . 21Storage Warranty Costs . 23Summary . 23Acknowledgements. 24Thin Client Computing . 24Citrix . 24Contact Information . 24Appendix A: ISE 230 Tested Configuration Details . 262 High-Performance 750-Seat VDI Reference Architecture powered by X-IO and Citrix

Executive SummaryThis document describes a fully-tested, benchmarked and proofed reference architecture for large-scale(750 user) virtual desktop infrastructure (VDI) implementations.The target audience for this document includes storage and virtualization administrators, consulting datacenter architects, field engineers, cloud architects, and desktop specialists who want to implement CitrixXenDesktop based virtual desktops. A working knowledge of Hyper-V, XenDesktop, server, storage,network and data center design is assumed but is not a prerequisite to read this document.VDI (Virtual Desktop Infrastructure) enables IT to deliver a secure, mobile, manageable and less costlyend user computing platform, at lower cost. That is the essential promise of VDI.However a large number of VDI projects fail to deliver on that promise, in most cases due to storageperformance bottlenecks. As a result: The VDI project cannot be completed, successfully at cost. The performance of the VDI environment is significantly worse than the performance of dedicatedhardware desktops. The cost of storage sufficient to equalize the VDI performance deficit makes the cost of the VDIproject greater than the cost of dedicated-hardware desktops. The storage performance bottlenecks that cripple so many VDI projects are the result, in somecases, of a failure to characterize the per-VM performance requirements for a VDI environment, butare more often the result of a damaging, single-minded focus on lower cost storage capacity ( /GB). In our view, it is absolutely critical that VDI projects balance both storage performance and storagecapacity in order to be successful.When per-VM storage performance requirements are taken into consideration along with per-VM storagecapacity requirements, during the planning phase of VDI projects:-Traditional enterprise storage array vendors are forced to over-provision, or short-stroke, storagecapacity to deliver enough storage performance for the required number of VMs, producing a veryexpensive solution.-Flash and solid state disk (SSD) array vendors wildly over-provision storage performance todeliver enough storage capacity to each VM.This reference architecture and its underlying results demonstrate, without a doubt, the suitability of anall-hard disk drive (HDD) ISE configuration without the need for higher cost all flash or hybrid arrays.3 High-Performance 750-Seat VDI Reference Architecture powered by X-IO and Citrix

Infrastructural ComponentsThe tested VDI reference architecture described in this document is based on the following: Citrix XenDesktop 7 Microsoft Windows Server 2012 R2 Hyper-V X-IO Intelligent Storage Elements (ISEs) and ISE Manager software Login VSICitrix XenDesktopCitrix XenDesktop delivers Windows apps and desktops as secure mobile services. With XenDesktop, ITcan mobilize the business, while reducing costs by centralizing control and security for intellectualproperty. Incorporating the full power of XenApp, XenDesktop can deliver full desktops or just the apps toany device. HDX technologies enable XenDesktop to deliver a native touch-enabled look-and-feel that isoptimized for the type of device, as well as the network.Citrix XenDesktop provides the best method available today to provision, host, manage and deliver virtualdesktops and applications. Citrix has the benefit of 25 years of advanced expertise in optimizing remoteprotocols and virtualizing user workloads.User experience is greatly enhanced with Citrix XenDesktop by providing high performance, 24x7 accessto all corporate apps and data, from anywhere, from any type of device (PC, Mac, Linux, Tablets, IOS,Android, etc.)For the IT department, Citrix centralizes management in dramatically more efficient ways. Rather thanmaintaining the legacy approach of installing and managing operating systems, applications, patches andcustomizations on every single device, Citrix allows a “One to Many” solution. A single desktop image canservice 100’s, 1000’s, or, 10’s of thousands of users providing a single point of update for the entire ITenvironment. At the same time, security and performance are greatly improved by keeping theapplications close to the data in the secure boundary of the data center.Microsoft Windows Server 2012 R2 Hyper-VHyper-V is an important part of Windows Server and provides a highly developed virtualization platformthat lets cloud providers provision for customers, and enterprises build their own private cloud of hosteddesktops. With Windows Server 2012 R2, customers get a compelling solution for core virtualizationscenarios; production server consolidation, dynamic data center, business continuity, test anddevelopment and most importantly, as it relates to this reference architecture, Virtual DesktopInfrastructure (VDI). Hyper-V provides customers with better flexibility with features like live migration andcluster shared volumes for storage flexibility. In Windows Server 2012 R2, Hyper-V also delivers greaterscalability and improved performance with support for dynamic memory and enhanced networkingsupport versus other competing hypervisors or previous versions of Hyper-V.Virtualization technologies help customers lower costs and deliver greater agility and economies of scale.Either as a stand-alone product or an integrated part of Windows Server, Hyper-V is the leadingvirtualization platform for today and the transformational opportunity with cloud computing, and hosteddesktops.4 High-Performance 750-Seat VDI Reference Architecture powered by X-IO and Citrix

The X-IO Intelligent Storage ElementX-IO’s Intelligent Storage Elements (ISEs) – both the 700 series hybrid storage arrays and the 200 seriesall-hard disk drive storage arrays – are frequently deployed for large-scale VDI projects because: ISEs offer predictably high levels of VDI performance. Performance is critical to the success of VDIimplementations. It impacts everything: user experience, application performance and the overall cost ofthe environment. ISEs scale linearly. The ISE can support thousands of virtual desktops with a “scale-out” architecturedesign that provides modular, linear scaling as needed for VDI deployments. ISEs offer industry-leading VDI price/performance. The ISE’s balance of performance, capacity andprice significantly improves the economics of large-scale virtual desktop environments, when comparedto conventional enterprise storage arrays. ISEs offer best-in-class reliability. ISEs are highly available, supporting today’s high-availabilityenterprise applications reliability demands with 99.999% uptime. Redundancy at every level of the ISEensures that if a hardware failure occurs, all application workloads will continue unaffected. X-IOTechnologies is so confident in the ISE’s technical and operational reliability that it warranties the ISEfor five years at no cost to its customers. ISEs are easy to deploy. Traditional storage systems are complex and difficult to troubleshoot, whenadjusting performance in VDI deployments. By contrast, the ISE Manager software, an intuitive storagemanagement environment that provides simple, user-friendly storage and integrated serveradministration, allows virtualization administrators to provision and troubleshoot the entire VDI “stack”through a single, easy-to-use graphical interface.Figure 1 – ISE Manager Storage View5 High-Performance 750-Seat VDI Reference Architecture powered by X-IO and Citrix

Login VSILogin VSI is and industry standard load testing tool for Virtual Desktop Infrastructures and Server BasedComputing environments.With Login VSI you can test the performance and scalability of Citrix XenDesktop and Citrix XenApp, bysimulating unique user workloads.With Login VSI you can: Benchmark different desktop, server and storage solutions and make the right decisions. Know how much hardware you need by right sizing your XenDesktop infrastructure. Gain insight in the maximum capacity by load testing your XenDesktop environment. Test the impact of changes on the performance before they are implemented.6 High-Performance 750-Seat VDI Reference Architecture powered by X-IO and Citrix

Reference Architecture ConfigurationThe system-under-test (SUT) was configured as described below.Physical Hardware Configuration One Hewlett-Packard C7000 blade enclosure with 16 BL460 G7 servers, Intel Xeon X5650 (2.67GHz, 6 cores), 192 GB memory, Windows Server 2012o Hyper-Threading was enabled in the BIOS of all 16 blade serverso Max Performance was enabled in the BIOS of all 16 blade serverso Two BL460 Servers were configured for Infrastructure Serviceso Three BL460 Servers were configured for Login VSI Client Workload Launcherso Eleven BL460 Servers were configured for hosted desktop virtualizationOne QLogic 5802v Fibre Channel switchTwo X-IO ISE 230 storage systemso Eleven write cache file LUNs; one LUN per XenDesktop blade servero Eleven user data LUNs, one LUN per XenDesktop blade serverFigure 2 shows the detailed topology of the reference architecture including the HP Blade enclosurewith server role allocations, SAN topology and X-IO ISE storage arrays.Figure 2 - Detailed Reference Architecture Configuration7 High-Performance 750-Seat VDI Reference Architecture powered by X-IO and Citrix

Virtual Machine ConfigurationInfrastructure Services: Windows Server 2012 Hyper-V Windows Server 2012 Virtual Machineso Active Directory, DHCP and DNS – 4 GB memory, 1 vCPUo Citrix Provisioning Services (PVS) 7.0 running SQL Server – 4 GB memory, 2 vCPUo Citrix Desktop Delivery Controller (DDC) running SQL Server – 8 GB memory, 2 vCPUo Citrix License server VMLogin VSI client launchers: Windows Server 2012 Hyper-V Windows 7 32-bit Virtual Machines – 1.5 GB memory, 1 vCPUXenDesktop Server: Windows Server 2012 R2 Hyper-V with MPIO feature added Windows 7 64-bit Virtual Machines – 1.5 GB memory, 1 vCPUVirtual Machine Configuration DiagramFigure 3 shows a topology of the tested environment for this reference architecture.The Citrix XenDesktop 7 infrastructure was configured on dedicated hosts. The infrastructure VMs werehosted on local server storage.Figure 3 - Test Environment overview of Citrix XenDesktopdeployment with Infrastructure components, WindowsServer 2012 R2 Hyper-V hosts and ISE volumes8 High-Performance 750-Seat VDI Reference Architecture powered by X-IO and Citrix

Citrix XenDesktop Storage RequirementsXenDesktop FlexCast Model has two different data types (write cache files and user data) with completelydifferent storage performance criteria.The following table details the tested ISE per-user storage configuration:Data TypePer user capacityWrite cache files8GB/user x 750 users (6TB total)User data41.7GB/user x 750 users (31.3TB total)Total49.7GB/user x 750 users (37.3TB)X-IO Tested Storage ConfigurationWindows Server 2012 R2 Configuration and Best PracticesThe Windows Server 2012 R2 Hyper-V role and MPIO feature were added using the following PowerShellcommand:# Install-WindowsFeature Hyper-V, Hyper-V-PowerShell, Hyper-V-Tools, Multipath-IOThe Windows Server power plan was set to “High performance” using the following PowerShellcommand:# POWERCFG.EXE /S SCHEME MINThe ISEs utilized the Windows Server 2012 R2 Multi-Path IO Microsoft Device Specific Module (MSDSM).Native DSM: Does not automatically claim enterprise class storage device LUNs, by default Uses a round-robin load balancing policy, by defaultThe tested MPIO configuration added “Xiotech ISE2400” to native DSM managed devices. MPIO Policywas set to Least Queue Depth (LQD).The following mpclaim commands were used, via PowerShell, to set all MPIO load balance policy for“XIOTECH ISE2400” for all LUNs.# List enterprise storage device for vendor stringmpclaim –e# List all MPIO disks controlled by MSDSM (native DSM)mpclaim –s –d# Add multipath support for device# Suppress rebootmpclaim –n –I –d “XIOTECH ISE2400”# Set load balance policy to LQD for all disks controlled by MSDSMmpclaim –l –m 49 High-Performance 750-Seat VDI Reference Architecture powered by X-IO and Citrix

# Restart server with reconfigured MPIO settingsshutdown –r –t 00Network ConfigurationNetwork 4 was configured to allow for Fibre connectivity from the HP blade servers. These wereconnected to an 8Gb QLogic 5802v. Four networks will be created:o Network 1 – Internal to HP Virtual Connect for PXE boot of VMs, allocated 5Gb bandwidtho Network 2 – Lab storage and management, 1Gb bandwidtho Network 3 – Production network for workload clients and XenDesktop VMs, 5Gb bandwidtho Network 4 – Fibre connection to ISE storage arrays, 8Gb bandwidthNote: There will be no HA or redundancy across the NICsFigure 4 illustrates the HP Blade Enclosure internal network design and external storage network.Figure 4 - Network Configuration10 High-Performance 750-Seat VDI Reference Architecture powered by X-IO and Citrix

Intelligent Storage Element Test ConfigurationTwo ISE 230 systems providing eighty (80) 900 GB, 10K RPM SAS hard disk drives rated at 57.6 TBusable capacity and 27,000 IOPS were used for this tested reference configuration. The ISEs providedthat performance and capacity in 6U of rack space, while consuming less than 1.5 kilowatts (kW) ofenergy. The ISEs were not tuned in any way before or during the tests described in the next section.There are four models in the ISE 200 series, each delivering real-world performance of 13,500 IOPS andbetween 9.6 and 38.4TB of usable capacity in each 3U array.Number ofISERated IOPSISE 210UseableCapacityISE 220UseableCapacityISE 230UseableCapacityISE 240UseableCapacity113,500 IOPS9.6 TB19.2 TB28.8 TB38.4 TB227,000 IOPS19.2 TB38.4 TB57.6 TB76.8 TBFor this reference architecture, the storage configuration must deliver a mix of aggregate performance ata minimum of 25,000 IOPS (30 IOPS per defined user VM) and useable capacity (8 GBs/user for writecache files and 41.7 GB/user for user data): easily within the rated performance and rated capacity of apair of ISE 230s.Fibre Channel Switch ConfigurationA single QLogic 5802v 8Gb Fibre Channel (FC) switch with 8 licensed ports was installed per establishedQLogic guidelines and user documentation. The Fibre Channel zoning was configured to be open.For smaller number of VDI users, it is possible to remove the Fibre Channel switch from the configurationcompletely and connect up to four servers directly to a single ISE via the server’s Fibre Channel host busadapter.For high availability configurations, a second QLogic 5802v 8Gb FC switch is required for VDIconfigurations requiring multiple ISE arrays and/or multiple servers.11 High-Performance 750-Seat VDI Reference Architecture powered by X-IO and Citrix

LUN ConfigurationIn the tables below, the ISEs are identified by unit letter (A or B), by use (L for write cache file and D foruser data) and blade number (blade 01 – 11). For example, “ISE-A-L01” denotes a LUN on ISE A, usedfor write cache, and mapped to Blade 1.ISE-A LUN Name Data Use RAID Type ISE-A-L01 Write cache ISE-A-L02 Write cache ISE-A-L03 ISE-A-L04 LUN Size ServerISE RAID-10 551 GB Blade 01 ISE RAID-10 552 GB Blade 02Write cache ISE RAID-10 553 GB Blade 03 Write cache ISE RAID-10 554 GB Blade 04ISE-A-L05 Write cache ISE RAID-10 555 GB Blade 05 ISE-A-L06 Write cache ISE RAID-10 556 GB Blade 06 ISE-A-D01 User data ISE RAID-5 3001 GB Blade 01 ISE-A-D02 User data ISE RAID-5 3002 GB Blade 02 ISE-A-D03 User data ISE RAID-5 3003 GB Blade 03 ISE-A-D04 User data ISE RAID-5 3004 GB Blade 04 ISE-A-D05 User data ISE RAID-5 3005 GB Blade 05ISE-BLUN NameData UseRAID TypeLUN SizeServerISE-B-L07Write cacheISE RAID-10557 GBBlade 07ISE-B-L08Write cacheISE RAID-10558 GBBlade 08ISE-B-L09Write cacheISE RAID-10559 GBBlade 09ISE-B-L10Write cacheISE RAID-10560 GBBlade 10ISE-B-L11Write cacheISE RAID-10561 GBBlade 11ISE-B-D06User dataISE RAID-53006 GBBlade 06ISE-B-D07User dataISE RAID-53007 GBBlade 07ISE-B-D08User dataISE RAID-53008 GBBlade 08ISE-B-D09User dataISE RAID-53009 GBBlade 09ISE-B-D10User dataISE RAID-53010 GBBlade 10ISE-B-D11User dataISE RAID-51244 GBBlade 1112 High-Performance 750-Seat VDI Reference Architecture powered by X-IO and Citrix

Login VSI ConfigurationLogin VSI is designed to perform benchmarks for VDI workloads through system saturation. Login VSIloads the system with simulated user workloads using well known desktop applications such as MicrosoftOffice, Internet Explorer and Adobe PDF reader.Login VSI is the de-facto industry standard benchmarking tool to test the performance and scalability ofcentralized Windows desktop environments. Login VSI is 100% vendor independent and is used to testvirtual desktop environments like Citrix XenDesktop.Heavy WorkloadThis workload is based on the standard Login VSI heavy knowledge worker workload, employing a mix ofMicrosoft Office applications, Internet Explorer, Adobe Acrobat (PDF) and FreeMind, a Java-basedmindmapping, note-taking and information management tool. Once a test session has been started, theworkload will repeat (loop) every 48 minutes. This loop is divided in segments, and each consecutiveLogin VSI user logon will start in a different segment. This ensures that all elements in the workload areequally used throughout the test. The heavy workload opens up to 5 applications simultaneously, withthe keyboard type rate set at 160ms for each character. During each loop the response time is measuredevery 3 - 4 minutes with only 2 minutes of idle time to simulate real‐world users.Each loop will open and use: Outlook -- browse messages. Internet Explorer -- browsing different webpages and a YouTube style video (720p or 1080p movietrailer) is opened three times in every loop. The heavy workload also begins by opening 4 instancesof Internet Explorer. These instances stay open throughout the workload loop. Word -- one instance to measure response time, one instance to review and edit a document. Doro PDF Printer & Acrobat Reader -- the Word document is printed and reviewed to PDF. Theheavy workload also begins by opening 2 instances of Adobe Reader, and increasing the number ofPDF printer actions in the workload. These instances stay open throughout the workload loop. Excel -- a very large randomized sheet is opened. PowerPoint -- a presentation is reviewed and edited. FreeMind – notes are taken.13 High-Performance 750-Seat VDI Reference Architecture powered by X-IO and Citrix

Test ResultsThis reference architecture was tested in November 2013, at the Citrix Labs in Santa Clara, California, toverify its ability to support at least 750 simultaneous VDI sessions/instances.Two sets of tests were performed in each pass of the testing process:Test 1 – Boot Storm Test: 750 user VMs must boot and register with a VSI broker within 10 minutes, withboot start occurring at the rate of 75 VMs per minuteTest 2 – Login VSI “Heavy” Workload Test: 750 user VMs must run the Login VSI “heavy” knowledgeworker workload, with average response time latency not exceeding 4 seconds (4000 milliseconds).The testing protocol for this reference architecture was as follows: After a test is performed, the response times are analyzed to calculate the maximum activesession/desktop capacity, and the maximum number of active sessions recorded as the VSImax1value. If the system-under-test is not saturated, and completed the full test without exceeding the averageresponse time latency threshold (in this test suite, 4000 milliseconds), the system-under-testreceives that tested VSImax rating, and a Pass. If the system is saturated, or the response time of the applications exceeds the average responsetime latency threshold (4000 milliseconds), the system does not receive a VSImax rating andreceives a Fail.Three test runs were performed on the configuration described in the previous section. Each test run wasmarked Pass; no test runs resulted in saturation or exceeded the response time latency threshold, and noconfiguration changes were made to the ISE storage between test passes. Server changes were madebetween test passes to increase the number of active VDI sessions achievable with the configuration.Detailed storage performance information was recorded, during each pass, with the ISE PerformanceAdapter for Windows Performance Monitor. This tool, included with ISE Manager, allows for extensiveperformance data capture and analysis using the Windows Performance Monitor Framework.1For more information, see tle VSImax14 High-Performance 750-Seat VDI Reference Architecture powered by X-IO and Citrix

ISE 230 Test ResultsBoot Storm TestFigure 5 is the main performance graph of both ISEs during the “Boot Storm” phase of the testing.Figure 5 - Boot Storm Read and Write IOPSDuring this phase, the ISE storage systems can be seen regularly servicing IO spikes over 20,000 IOPS,with peaks of over 30,000 IOPS. Write IOs comprise the vast majority of the workload, which the ISEarrays handled easily. At peak times, the pair of ISEs delivered approximately 375 IOPS per hard diskdrive.15 High-Performance 750-Seat VDI Reference Architecture powered by X-IO and Citrix

In figure 6, most of the Read Latencies during the boot storm test were below 3ms. Latency below 10 msis considered good, in a VDI environment, and the ISEs lower latencies indicate that the system isperformance-optimized for the workload.Figure 6 - Boot Storm Read LatencyIn figure 7, the Write Latencies during the Boot Storm test were below 1ms. Again, the ISEs were easilyable to handle the workload.Figure 7 - Boot Storm Write Latency16 High-Performance 750-Seat VDI Reference Architecture powered by X-IO and Citrix

Login VSI Heavy Workload TestIn Figure 8, the vast majority of the workload is shown to be heavily write-oriented (or “write-intensive”) innature. Writes peaked, during this test, at over 7,000 per second, and were routinely at or above the6000 IOPS level.Figure 8 - User Load Read and Write Total IOPSIn Figure 9, 95% of the read latency values were below 3 milliseconds, with a single observed high valueof 10ms, indicating clearly that the ISEs were having no difficulty servicing the read-oriented portion of theworkload.Figure 9 - User Load Average Read Latency17 High-Performance 750-Seat VDI Reference Architecture powered by X-IO and Citrix

Figure 10 shows write latency on the ISE arrays under test. Any latency values below 5-10 ms areconsidered good. The write latency results below are excellent and show the ISE arrays are able toservice these workloads with no signs of reaching a performance limit.Figure 10 - User Load Average Write LatencyThe data recorded from the ISE arrays above demonstrate a high performance, write-intensive workload.The ISE arrays easily handle this intensive workload as evident by the sub-millisecond latency responsetimes. This is very important to consider as many high performance storage products on the market areoptimized for random, read-intensive operations, through the use of a flash-enabled read-cache, forexample.Login VSI Test ResultsThe Login VSI VSImax chart below documents the actual number of maximum virtual desktop sessionssupported by the configuration.18 High-Performance 750-Seat VDI Reference Architecture powered by X-IO and Citrix

Figure 11 - VSImaxFigure 11 shows Login VSI v4.0 VSImax is not reached, setting VSImax for this test at 759. With ReadLatencies mostly at 3ms and Write Latencies at 1ms, the ISE arrays are carrying the workload of the 750XenDesktop VMs without issue.19 High-Performance 750-Seat VDI Reference Architecture powered by X-IO and Citrix

Total Cost of Ownership (TCO) AnalysisThe test results for this reference architecture demonstrate clearly that the configuration is a highperformance solution for large-scale VDI workload requirements. The documented reference architectureis also a superior economic alternative, as this total cost of ownership (TCO) analysis demonstrates.The following analysis shows a cost savings of 91,400 of the tested ISE solution in this referencearchitecture as compared to a similarly performant solution based on traditional enterprise storage arrays.TCO versus Cost of AcquisitionThe initial purchase price of any VDI system is just the beginning of the expenses that must be budgetedand paid for. System requirements for rack space, power requirements and consumption, heatgeneration, failure remediation, and more are all factors that increase the total cost of the solution whencalculated over the lifetime of the hardware (5 years).Sizing Storage for Actual Performance RequirementsThe Login VSI Heavy User Workload test used to proof this reference architecture requires 6,000 to 8,000write IOPS, while the Boot Storm test requires more than 25,000 write IOPS. Selecting storage based onsteady-state workload, rather than boot storm workload, almost always results in under-performant VDIinfrastructure and dissatisfied VDI users.To illustrate this point: traditional enterprise storage vendors with products that utilize 10K RPM enterpriseSAS hard disk drives (the same drives used in the ISE 200 series arrays) typically rate their drives at 125IOPS per drive. To deliver 8,000 IOPS for the Login VSI heavy user workload test would require 64 ofthese drives. To deliver Boot Storm ( 25,000 IOPS) performance would require over 200 of these drives.Figure 12 - Number of Disks Required for Login VSI WorkloadsPurchasing a storage system based on steady-state user workloa

Feb 20, 2014 · 7 High -Performance 750 Seat VDI Reference Architecture powered by X IO and Citrix Reference Architecture Configuration The system-under-test (SUT) was configured as described below. Physical Hardware Configuration One Hewlett-Packard C7000 blade e