Dell EMC Ready Architectures for VDI

Transcription

1 Dell EMC Ready Architectures for VDI Designs for Citrix Virtual Apps and Desktops on Dell EMC XC Family October 2018 H Validation Guide Abstract This validation guide describes the architecture and performance of the integration of Citrix Virtual Apps and Desktops components for virtual desktop infrastructure (VDI) and hosted shared desktops on Dell EMC XC Family devices in a VMware vsphere environment. Dell EMC Solutions

2 Copyright 2018 Dell Inc. or its subsidiaries. All rights reserved. Published October 2018 Dell believes the information in this publication is accurate as of its publication date. The information is subject to change without notice. THE INFORMATION IN THIS PUBLICATION IS PROVIDED AS-IS. DELL MAKES NO REPRESENTATIONS OR WARRANTIES OF ANY KIND WITH RESPECT TO THE INFORMATION IN THIS PUBLICATION, AND SPECIFICALLY DISCLAIMS IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. USE, COPYING, AND DISTRIBUTION OF ANY DELL SOFTWARE DESCRIBED IN THIS PUBLICATION REQUIRES AN APPLICABLE SOFTWARE LICENSE. Dell, EMC, and other trademarks are trademarks of Dell Inc. or its subsidiaries. Other trademarks may be the property of their respective owners. Published in the USA. Dell EMC Hopkinton, Massachusetts In North America Dell EMC Ready Architectures for VDI

3 CONTENTS Chapter 1 Executive Summary 5 Document purpose... 6 Audience... 6 We value your feedback... 6 Chapter 2 Test Environment Configuration and Best Practices 7 Validated hardware resources... 8 Enterprise hardware... 8 Graphics hardware...8 Network hardware...8 Validated software resources... 9 Validated system version matrix... 9 Virtual networking configuration Management server infrastructure...10 NVIDIA GRID License Server...11 SQL Server databases DNS High availability...11 Citrix Virtual Apps and Desktops architecture Chapter 3 Solution Performance and Testing 13 Testing process...14 Resource monitoring Load generation Profiles and workloads Citrix Virtual Apps and Desktops Machine Creation Services Desktop VM test configurations Test results and analysis...17 B5 configuration...19 C7 configuration Chapter 4 Conclusion 51 Density recommendations Summary Chapter 5 References 53 Dell EMC documentation...54 VMware documentation Citrix resources Dell EMC Ready Architectures for VDI 3

4 CONTENTS 4 Dell EMC Ready Architectures for VDI

5 CHAPTER 1 Executive Summary This chapter presents the following topics: Document purpose...6 Audience... 6 We value your feedback... 6 Executive Summary 5

6 Executive Summary Document purpose Audience We value your feedback This validation guide details the architecture, components, testing methods, and test results for Dell EMC XC Family devices with Citrix Virtual Apps and Desktops. It includes the test environment configuration and best practices for systems that have undergone testing. This guide is intended for architects, developers, and technical administrators of IT environments. It provides an in-depth explanation of the testing methodology and basis for VDI densities. It also validates the value of the Dell EMC Ready Architectures for VDI that deliver Microsoft Windows virtual desktops to users of Citrix Virtual Apps and Desktops VDI components on XC Family devices. Dell EMC and the authors of this document welcome your feedback on the solution and the solution documentation. Contact Dell EMC Solutions team with your comments. Authors: Dell EMC Ready Architectures for VDI Engineering Team, Donna Renfro 6 Dell EMC Ready Architectures for VDI

7 CHAPTER 2 Test Environment Configuration and Best Practices This chapter presents the following topics: Validated hardware resources... 8 Validated software resources...9 Validated system version matrix...9 Virtual networking configuration...10 Management server infrastructure High availability Citrix Virtual Apps and Desktops architecture...12 Test Environment Configuration and Best Practices 7

8 Test Environment Configuration and Best Practices Validated hardware resources Enterprise hardware Dell EMC validated the solution with the specific hardware resources listed in this section. We used the Dell EMC XC Family XC740xd-24 device with the components listed in the following table. We have designated the configurations as B5 and C7, which are referenced throughout the document. Table 1 Validated hardware configurations Configur ation Enterprise platform CPU Memory RAID controller HD configuration Network B5 XC740xd Gold (14- core 2.2 GHz) 384 2,400 MT/s HBA x 240 GB M.2 2 x 960 GB SSD 4 x 1.8 TB HDD 2 x Mellanox Connect X-4 LX 25 GbE SFP Rack NDC C7 XC740xd Gold (20- core 2.0 GHz) 768 2,666 MT/s HBA x 240 GB M.2 2 x 960 GB SSD Intel X710 rndc 6 x 1.8 TB HDD Graphics hardware We used the following NVIDIA GPU hardware in our tests for graphics-intensive workloads: NVIDIA Tesla M10 A dual-slot 10.5-inch PCI Express Gen3 graphics card featuring four mid-range 8 GB NVIDIA Maxwell GPUs and a total of 32 GB GDDR5 memory per card NVIDIA Pascal P40 A dual-slot 10.5-inch PCI Express Gen3 graphics card featuring a single high-end NVIDIA Pascal GPU and a total of 24 GB GDDR5 memory per card Network hardware We used the following network hardware in our test environment: Dell Networking S3048 (1 GbE ToR switch) A low-latency top-of-rack (ToR) switch that features 48 x 1 GbE and 4 x 10 GbE ports, a dense 1U design, and up to 260 Gbps switch fabric capacity Dell Networking S4048 (10 GbE ToR switch) A high-density, ultra-low-latency ToR switch that features 48 x 10 GbE SFP+ and 6 x 40 GbE ports and up to 720 Gbps switch fabric capacity 8 Dell EMC Ready Architectures for VDI

9 Test Environment Configuration and Best Practices Validated software resources Dell EMC validated this solution with the software components that are listed in the following table. Table 2 Validated software components Component Hypervisor Broker technology Description/Version VMware vsphere ESXi 6.5, Microsoft Hyper-V Server 2012 R2 and 2016 Citrix Virtual Apps and Desktops version 7.15 LTSR Broker database Microsoft SQL Server 2016 Management VM operating system Virtual desktop operating system Microsoft Windows Server 2016 (DCC, StoreFront, and database) Microsoft Windows 10 Enterprise Office application suite Microsoft Office Professional 2016 Login VSI test suite Version 4.1 Platform Nutanix AOS version NVIDIA GRID software (for graphics testing) 6.2 Validated system version matrix Dell EMC validated this solution using the system versions listed in the following table. Table 3 Version matrix for tested system Server configururation Hypervisor Hypervisor version Hypervisor build Hypervisor patches Bios AOS version Windows 10 version Windows 10 patches Nvidia vgpu version B5 + 3 x P40 ESXi 6.5 U KB C7 ESXi 6.5 U C7 Hyper-V 2012 R2 6.3 build 9600 C7 Hyper-V build LTSB KB KB C7 + 3 x M60 C7 + 3 x P40 ESXi 6.5 U ESXi 6.5 U KB Validated software resources 9

10 Test Environment Configuration and Best Practices Virtual networking configuration The network configuration for the Dell EMC XC Family devices uses a 10 Gb converged infrastructure model. All required VLANs traverse two 10 Gb NICs configured in an active/active team. For larger scaling, we recommend that you separate the infrastructure management virtual machines (VMs) from the compute VMs to aid in predictable compute host scaling. We used the following VLAN configurations for the compute hosts, management hosts, and idrac in this solution model: Compute hosts Management VLAN: Configured for hypervisor infrastructure traffic L3 routed by using the spine layer Live Migration VLAN: Configured for Live Migration traffic L2 switched by using the leaf layer VDI VLAN: Configured for VDI session traffic L3 routed by using the spine layer Management hosts Management VLAN: Configured for hypervisor management traffic L3 routed by using the spine layer Live Migration VLAN: Configured for Live Migration traffic L2 switched by using the leaf layer VDI Management VLAN: Configured for VDI infrastructure traffic L3 routed by using the spine layer VLAN idrac: Configured for all hardware management traffic L3 routed by using the spine layer Management server infrastructure The following table lists the sizing requirements for the management server components. Table 4 Sizing for XC Family devices Component vcpus RAM (GB) NICs Operating system + data vdisk (GB) Tier 2 volume (GB) VMware vcenter Appliance Platform Services Controller Desktop Delivery Controller and License Server StoreFront SQL Server (VMDK) File server ,048 (VMDK) 10 Dell EMC Ready Architectures for VDI

11 Test Environment Configuration and Best Practices Table 4 Sizing for XC Family devices (continued) Component vcpus RAM (GB) NICs Operating system + data vdisk (GB) Tier 2 volume (GB) Nutanix CVM SCVMM RDSH VM NVIDIA GRID License Server SQL Server databases When using NVIDIA vgpu cards, graphics-enabled VMs must obtain a license from a GRID License Server on your network to be entitled for vgpu. We installed the GRID License Server software on a system running a Windows 2012 R2 operating system to test vgpu configurations. We made the following changes to the GRID License Server to address licensing requirements: Used a reserved fixed IP address Configured a single MAC address Applied time synchronization to all hosts on the same network During validation, a single dedicated SQL Server 2016 VM hosted the VMware databases in the management layer. We separated SQL data, logs, and tempdb into their respective volumes, and created a single database for Desktop Delivery Controller and License Server. We adhered to VMware best practices for this testing, including alignment of disks to be used by SQL Server with a 1,024 KB offset and formatted with a 64 KB file allocation unit size (data, logs, and tempdb). DNS DNS is the basis for Microsoft Active Directory and also controls access to various software components for VMware services. All hosts, VMs, and consumable software components must have a presence in DNS. We used a dynamic namespace integrated with Active Directory and adhered to Microsoft best practices. High availability Although we did not enable high availability (HA) during the validation that is documented in this guide, we strongly recommend that HA be factored into any VDI design and deployment. This process involves following the N+1 model with redundancy at both the hardware and software layers. The design guide for this architecture provides additional recommendations for HA. NVIDIA GRID License Server 11

12 Test Environment Configuration and Best Practices Citrix Virtual Apps and Desktops architecture The following figure shows the Citrix Virtual Apps and Desktops communication flow. Figure 1 Citrix Virtual Apps and Desktops architecture 12 Dell EMC Ready Architectures for VDI

13 CHAPTER 3 Solution Performance and Testing This chapter presents the following topics: Testing process Test results and analysis Solution Performance and Testing 13

14 Testing process Resource monitoring To ensure the optimal combination of end-user experience (EUE) and cost-per-user, we conducted performance analysis and characterization testing on this solution using the Login VSI load-generation tool. Login VSI is a carefully designed, holistic methodology that monitors both hardware resource utilization parameters and EUE during load-testing. We tested each user load against four runs: a pilot run to validate that the infrastructure was functioning and valid data could be captured, and three subsequent runs to enable data correlation. During testing, while the environment was under load, we logged in to a session and completed tasks that correspond to the user workload. While this test is subjective, it helps to provide a better understanding of the EUE in the desktop sessions, particularly under high load. It also helps to ensure reliable data gathering. To ensure that the user experience was not compromised, we monitored the following important resources: Compute host servers VMware vcenter (for VMware vsphere-based solutions) or Microsoft Performance Monitor (for Hyper-V-based solutions) gathers key data (CPU, memory, disk and network usage) from each of the compute hosts during each test run. This data is exported to.csv files for single hosts, and then consolidated to show data from all hosts. While the report does not include specific performance metrics for the management host servers, these servers are monitored during testing to ensure that they are performing at an expected level with no bottlenecks. Hardware resources Resource contention, which occurs when hardware resources have been exhausted, can cause poor EUE. We monitored the relevant resource utilization parameters and applied relatively conservative thresholds, as shown in the following table. Thresholds are carefully selected to deliver an optimal combination of good EUE and cost-per-user while also providing burst capacity for seasonal or intermittent spikes in usage. Table 5 Parameter pass/fail thresholds Parameter Physical host CPU utilization Pass/fail threshold 85% a Physical host memory utilization 85% Network throughput 85% Storage I/O latency 20 ms a. The Ready Solutions for VDI team recommends that average CPU utilization not exceed 85% in a production environment. A 5% margin of error was allocated for this validation effort. Therefore, you will see CPU utilization that exceeds our recommended percentage. Given the nature of LoginVSI testing, this is a reasonable exception for determining our sizing guidance. GPU resources vsphere Client monitoring collects data about the GPU resource use from a script that is run on ESXi 6.5 and later hosts. The script runs for the 14 Dell EMC Ready Architectures for VDI

15 duration of the test and contains NVIDIA System Management Interface commands. The commands query each GPU and log the GPU processor, temperature, and memory use to a.csv file. Load generation Login VSI from Login VSI, Inc. is the industry-standard tool for testing VDI environments and Remote Desktop Session Host (RDSH) environments. Login VSI installs a standard collection of desktop application software (for example, Microsoft Office, Adobe Acrobat Reader) on each VDI desktop. It then uses launcher systems to connect a specified number of users to available desktops within the environment. When the user is connected, a logon script starts the workload, configures the user environment, and starts the test script. Each launcher system can launch connections to a number of VDI desktops (target machines). A centralized management console configures and manages the launchers and the Login VSI environment. In addition, we used the following login and boot paradigm: Users were logged in within a login timeframe of 1 hour, except when testing lowdensity solutions such as GPU/graphic-based configurations, in which users were logged in every 10 to 15 seconds. All desktops were started before users logged in. All desktops ran an industry-standard anti-virus solution. Windows 10 machines used Windows Defender. Profiles and workloads Machine profiles and user workloads determine the density numbers that the solution can support. Each profile and workload is bound by specific metrics and capabilities, with two targeted at graphics-intensive use cases. Profiles and workloads are defined as follows: Profile The configuration of the virtual desktop; the number of vcpus and the amount of RAM that is configured on the desktop and available to the user Workload The set of applications that is used We load-tested each profile by using a workload that is representative of the profile. The following table describes each use case. Table 6 Virtual desktop profiles and workloads Profile name/ workload Task worker Knowledge worker Workload description The least intensive of the standard workloads. This workload primarily runs Microsoft Excel and Microsoft Internet Explorer, with some minimal Microsoft Word activity, as well as Microsoft Outlook, Adobe, and copy and zip actions. The applications are started and stopped infrequently, which results in lower CPU, memory, and disk I/O usage. Designed for virtual machines with 2 vcpus. This workload includes the following activities: Outlook: Browse messages. Internet Explorer: Browse websites and open a YouTube style video (480p movie trailer) three times in every loop. Word: Initiate one instance to measure response time and another to review and edit a document. Load generation 15

16 Table 6 Virtual desktop profiles and workloads (continued) Profile name/ workload Workload description Doro PDF Printer and Acrobat Reader: Print a Word document and export it to PDF. Excel: Open a large randomized sheet. PowerPoint: Review and edit a presentation. FreeMind: Run a Java-based Mind Mapping application. Other: Perform various copy and zip actions. Power worker The most intensive of the standard workloads. The following activities are performed with this workload: Begin by opening four instances of Internet Explorer and two instances of Adobe Reader, which remain open throughout the workload. Perform more PDF printer actions than in the other workloads. Watch a 720p and a 1080p video. Reduce the idle time to two minutes. Perform various copy and zip actions. Graphics performance configuration/ multimedia A workload that is designed to heavily stress the CPU when using software graphics acceleration. GPU-accelerated computing offloads the most compute-intensive sections of an application to the GPU while the CPU processes the remaining code. This modified workload uses the following applications for its GPU/CPU-intensive operations: Adobe Acrobat Google Chrome Google Earth Microsoft Excel HTML5 3D spinning balls Internet Explorer MP3 Microsoft Outlook Microsoft PowerPoint Microsoft Word Streaming video Citrix Virtual Apps and Desktops Machine Creation Services For this validation, we used the Virtual Apps and Desktops Machine Creation Services (MCS) linked clones provisioning method. MCS is a collection of services that work together to create virtual desktops from a master image on demand, optimizing storage utilization and providing a pristine virtual machine to users each time they log on. 16 Dell EMC Ready Architectures for VDI

17 Desktop VM test configurations The following table summarizes the compute VM configurations for the profiles and workloads that we tested. Table 7 Desktop VM specifications User profile vcpus ESXi configured memory ESXi reserved memory Screen resolution Operating system Task worker 2 a 2 GB 1 GB 1280 x 720 Windows 10 Enterprise 64-bit Knowledge worker 2 3 GB 1.5 GB 1920 x 1080 Windows 10 Enterprise 64-bit Power worker 2 4 GB 2 GB 1920 x 1080 Windows 10 Enterprise 64-bit Multimedia 4 8 GB 8 GB 1920 x 1080 Windows 10 Enterprise 64-bit a. Dell EMC has validated the LoginVSI Task worker workload with two vcpus assigned per VM, although LoginVSI lists the typical VM vcpu profile for this workload as being a single vcpu. Dell EMC diverges from this definition to deliver virtual desktops with great user experience. Increasing the vcpu count to 2 in the vcpu profile associated with the Task worker workload does have a minor impact on densities but generates improved user experience in return. Test results and analysis We used the Login VSI test suite to simulate the user experience for several profile types under the typical workload for that type. The following table summarizes the test results for the compute hosts using the various workloads and configurations. Note Dell EMC is aware of the vulnerabilities, known as Meltdown, Spectre, and Foreshadow/L1TF, which affect many modern microprocessors. Ensure that you read the information in the following links: Consider this information in combination with the version information in Table 3 on page 9 to understand the vulnerability mitigation status of the environment we used to derive the test results shown in the following table. Table 8 Test results summary for VMware ESXi Server configuration Workload User density Avg CPU Avg active memory Avg IOPS per user B5 + 3 x P40 Multimedia (Virtual PC: P40-2B) 36 72% 321 GB 15 C7 Task worker % 427 GB 3.3 C7 Knowledge worker % 527 GB 4.2 C7 Power worker % 561 GB 3.5 C7 + 3 x M60 Power worker (Virtual Workstation: M60-1Q) 48 38% 134 GB 9.1 Desktop VM test configurations 17

18 Table 8 Test results summary for VMware ESXi (continued) Server configuration Workload User density Avg CPU Avg active memory Avg IOPS per user C7 + 3 x P40 Multimedia (Virtual PC: P40-1B) % 609 GB 6.4 Table 9 Test results summary for Microsoft Hyper-V Server configuration Workload User density Avg CPU Avg active memory Avg IOPS per user C7 - Hyper-V 2016 (Shared Desktop) Task worker % 338 GB 0.44 C7 - Hyper-V 2016 Knowledge worker % 522 GB 2.27 C7 - Hyper-V 2012 R2 Task worker % 370 GB 2.31 C7 - Hyper-V 2012 R2 Knowledge worker % 400 GB 1.89 C7 - Hyper-V 2012 R2 Power worker % 354 GB 1.68 The table headings are defined as follows: User density The number of users per compute host that successfully completed the workload test within the acceptable resource limits for the host. For clusters, this reflects the average of the density achieved for all compute hosts in the cluster. Avg CPU The average CPU usage over the steady-state period. For clusters, this represents the combined average CPU usage of all compute hosts. On the latest Intel processors, the ESXi host CPU metrics exceed the rated 100 percent for the host if Turbo Boost is enabled, which is the default setting. An additional 35 percent of CPU is available from the Turbo Boost feature, but this additional CPU headroom is not reflected in the VMware vsphere metrics where the performance data is gathered. Therefore, CPU usage for ESXi hosts is adjusted and each CPU graph includes a line indicating the potential performance headroom that is provided by Turbo boost. Avg active memory For ESXi hosts, the amount of memory that is actively used, as estimated by the VMkernel based on recently touched memory pages. For clusters, this is the average amount of guest physical memory that is actively used across all compute hosts over the steady-state period. Avg IOPS per user IOPS calculated from the average disk IOPS over the steady state period divided by the number of users. 18 Dell EMC Ready Architectures for VDI

19 B5 configuration We performed the following multimedia performance testing on the XC740xd system in the B5 configuration as described in Validated hardware resources on page 8. Multimedia workload (P40-1B), 36 vgpu users, ESXi 6.5 U2, XenDesktop 7.15 MCS We ran the following tests on this workload. CPU We populated the GPU-enabled compute host with 36 vgpu-enabled virtual machines (VMs) and used the NVIDIA P40-1B profile. With all user VMs powered on and before we started the test, the CPU usage was approximately 9 percent. The following figure shows the performance data for 36 user sessions on the management and GPU-enabled compute hosts. The CPU reached a steady-state average of 72 percent during the test cycle when all users were logged in to the GPUenabled compute host. GPU usage The GPU metrics came from the vsphere Client. As shown in the following figure, the GPU usage during the steady-state period averaged approximately 38.1 percent and reached a peak usage of percent with the multimedia workload. B5 configuration 19

20 Memory With regard to memory consumption for this test run, there were no constraints on the management or GPU-enabled compute hosts. Of a total of 384 GB available memory per node, the GPU compute host reached a maximum memory consumption of 336 GB with active memory usage reaching a maximum of 321 GB during the steady-state phase, as shown in the following figure. Each user session consumed 9.3 GB of memory and 8.9 GB of active memory after accounting for Nutanix Controller VM (CVM) and management memory usage. There were no variations in memory usage throughout the test, because all vgpu-enabled VM memory was reserved. No memory ballooning or swapping occurred on either host. 20 Dell EMC Ready Architectures for VDI

21 Network usage Network bandwidth, with a steady-state peak of approximately 136 Mbps, was not an issue on this test run. The busiest period for network traffic was during the logon phase. The compute and GPU hosts reached a peak of 303 Mbps at the start of steady state. IOPS The IOPS graphs and IOPS numbers came from the Nutanix Prism web console. The graphs clearly display the boot storm phase, the initial logon of the desktops, the steady state, and then the logoff phase. The graphs show IOPS data for the cluster and the GPU host. The following figure shows that the cluster IOPS reached a maximum of 4,306 disk IOPS during the boot storm and averaged 539 disk IOPS during steady state. Based on these numbers, each user session generated 15 disk IOPS during steady state. B5 configuration 21

22 The following figure shows that cluster controller IOPS reached a maximum of 7,975 disk IOPS during the boot storm and averaged 510 disk IOPS during steady state. Based on these numbers, each user session generated 14 IOPS in steady state. GPU host disk IOPS The GPU host reached a maximum of 3,152 disk IOPS during the boot storm and averaged 358 disk IOPS during steady state, as shown in the following figure. Based on these numbers, each user session generated 10 IOPS in steady state. 22 Dell EMC Ready Architectures for VDI

23 Latency The latency graphs and latency numbers came from the Nutanix Prism web console. As shown in the following figure, they clearly display the initial logon of the desktops, the steady state, the logoff phase, and finally the re-creation of the instant clones. The graphs show latency data for the cluster and the GPU host. The cluster latency reached a maximum latency of 7 ms during the boot storm and averaged 0.35 ms during steady state. The GPU host reached a maximum latency of 6.5 ms during the boot storm and averaged 0.37 ms during steady state. B5 configuration 23

24 User Experience The following figure shows that the user experience score did not reach the Login VSI maximum for this test. When we manually interacted with the sessions during steady state, the mouse response, window movement, and video playback was good. The baseline performance of 788 indicates that the user experience for this test run was good. The index average reached 1,045, which is well below the threshold of 1, Dell EMC Ready Architectures for VDI

25 Notes During the re-creation of the VMs, there was little network traffic on the compute/gpu host because the VMs were created on the other two hosts, and then migrated onto the compute/gpu host. No disk latency issues occurred during testing. B5 configuration 25

26 C7 configuration We ran the following multimedia performance testing on the XC740xd system in the C7 configuration as described in Validated hardware resources on page 8. Notes We deployed Virtual Apps and Desktops and vsphere management roles within the cluster on a single host that also hosted desktops. This optional configuration is beneficial for POCs or small deployments looking to maximize user density. We allocated 12 vcpus and 32 GB of memory to the Nutanix Controller VM CVM) when we configured the Nutanix cluster. We set the Nvidia Pascal P40 GPU accelerator card with vgpu scheduling policy to Fixed Share Scheduler. Task Worker, 560 users, ESXi 6.5, XenDesktop 7.15 MCS We ran the following tests on this workload. CPU usage For this test, we provisioned 190 VMs on each compute host and 180 VMs on the management host along with the XenDesktop management roles. The peak CPU usage was 94 percent on one node at the end of the logon phase, while the steady-state average CPU usage was 77 percent, as indicated in the following figure. Consumed memory The average consumed memory per host during the test run was 427 GB in the steady-state phase. The peak memory usage was 430 GB on one host during steady state. There was no ballooning or swapping of VM memory during the test run. Each user session consumed 2.07 GB of memory after accounting for CVM and management memory usage. The following figure shows the consumed memory during our testing. 26 Dell EMC Ready Architectures for VDI

27 Active memory The average active memory use during steady state was 180 GB, while the peak was 195 GB on one host during logoff phase. After accounting for VM and management memory usage, each user session used 0.78 GB of active memory during steady state. The following figure shows the active memory use during our testing. Network usage The steady-state average network usage was 667 Mbps. The peak was 2,140 Mbps during the boot storm phase, as shown in the following figure. C7 configuration 27

28 IOPS The peak cluster IOPS for the test run was 6,850 IOPS during the boot storm phase, while the average in steady state was 1,850 IOPS, as shown in the following figure. Based on these numbers, each user session generated 3.3 IOPS in steady state. I/O latency The peak I/O latency was 1.8 ms during the boot storm. The average I/O latency during steady state was 0.4 ms. The following figure clearly shows a steady low level of I/O latency throughout the test run. 28 Dell EMC Ready Architectures for VDI

29 User experience The baseline performance of 842 indicates that the user experience for this test run was good, as shown in the following figure. The index average reached 982, which is well below the threshold of 1,843. C7 configuration 29

30 Knowledge Worker, 475 users, ESXi 6.5, XenDesktop 7.15 MCS We ran the following tests on this workload. CPU usage In this workload test run, the compute hosts each had 160 user sessions, while the management host had 155 sessions in addition to the XenDesktop management VMs. The peak CPU usage was 99 percent on one host during the logon phase, while the steady-state average was 83 percent across all hosts. The relatively low steady-state CPU usage was essential to avoid CPU scheduling problems that would have placed many of the desktops into the ready-waiting state. The average CVM CPU usage during steady state was 7.8 percent, while the management VMs altogether used 1.2 percent. The following figure shows the CPU usage during this test. Consumed memory As shown in the following figure, the memory consumption averaged 527 GB in steady state across all hosts. The peak usage on any host was 532 GB during the steadystate phase. The peak usage was 69 percent, well below the 85 percent threshold, while the steady-state average usage for all hosts was 68.6 percent. There was no swapping or ballooning during the test run. After accounting for CVM and management VM memory consumption, each desktop consumed 3.06 GB, which was slightly more than the granted 3.0 GB. The CVM on each host consumed its full 32 GB of reserved memory throughout the test run. 30 Dell EMC Ready Architectures for VDI

31 Active memory During the test run, active memory peaked at 226 GB on the management host and averaged 204 GB during steady state. Each desktop accounted for 1.06 GB of active memory usage after deducting CVM and management VM active memory. The CVM on each host used a full 32 GB of active memory throughout the test run. Network usage Network usage peaked at 1,709 Mbps during boot storm on one host, and the average network usage for all hosts was 984 Mbps during steady state. Each desktop produced network throughput of 6.32 Mbps in steady state. C7 configuration 31

32 IOPS The peak cluster IOPS for the test run was 7,589 IOPS at the end of the boot storm, while the average in steady state was 1,997 IOPS. Based on these numbers, each user session generated 4.20 IOPS in steady state. I/O latency The peak cluster I/O latency was 0.9 ms during the boot storm. The average cluster I/O latency during steady state was 0.4 ms. The highest I/O latency on any host was 1.4 ms during the boot storm. The chart clearly shows a very steady and very low level of I/O latency throughout the test run. 32 Dell EMC Ready Architectures for VDI

33 User experience The baseline performance of 851 indicates that the user experience for this test run was good. The index average reached 1,524, which was well below the VSI maximum threshold of 1,851. C7 configuration 33

34 Power Worker, 390 users, ESXi 6.5, XenDesktop 7.15 MCS We ran the following tests on this workload. CPU usage In this workload test run, each compute host had 130 user sessions, while the designated management host had the full set of management VMs plus 130 desktops. The peak CPU usage was 99 percent on one host during the logon phase, while the steady-state average was 87 percent across all hosts. The relatively low steady-state CPU usage was essential to avoid CPU scheduling problems that would have placed many of the desktops into the ready-waiting state. The CVMs on each host averaged 7.5 percent CPU usage during steady state. The management VMs used only 1.4 percent CPU on the management host in steady state. Provisioning 390 power worker desktops with XenDesktop MCS Fast Cloning on vsphere 6.5 U1 took only 11 minutes 23 seconds, not counting time required to power on the entire complement of desktops, which was 40 minutes once the desktops were added to a desktop group. Of the total provisioning time, 5 min 37 seconds was required to prepare the base image, and the remaining time, 5 minutes 46 seconds, was for creating the cloned VMs. Consumed memory The memory consumption averaged 561 GB in steady state across all hosts, and the peak usage on any host was 584 GB during the steady-state phase. The peak usage was 76 percent, which was well below the 85 percent threshold. The steady-state average usage for all hosts was 73 percent. There was no swapping or ballooning during the test run and each desktop consumed 4.01 GB, which was slightly more than the granted 4.0 GB, after accounting for CVM and management VM memory consumption. 34 Dell EMC Ready Architectures for VDI

35 Active memory Active memory usage reached a maximum of 196 GB on the management host during steady state, and the average steady-state usage was 184 GB. Each desktop used 1.15 GB of active memory after deducting for CVM and management VM usage. The CVM used its full 32 GB of active memory throughout the test run. Network usage Network usage peaked at 1,495 Mbps on one host during the logon phase, and the average network usage for all hosts during steady state was 1,205 Mbps. Each desktop accounted for 9.27 Mbps in steady state. C7 configuration 35

36 IOPS The peak cluster IOPS for the test run was 6,885 IOPS during logon, while the average in steady state was 1,381 IOPS. Based on these numbers, each user session generated 3.54 IOPS during steady state I/O latency The peak cluster I/O latency was 0.8 ms during the boot storm, while the peak on any host was 1.0 ms. The average cluster I/O latency during steady state was 0.6 ms. The chart clearly shows a very steady and very low level of IO latency throughout the test run. 36 Dell EMC Ready Architectures for VDI

37 User experience The baseline performance of 834 indicates that the user experience for this test run was good. The index average reached 1,175, which was well below the threshold of 1,834. Although the difference between the VSI maximum and the average would seem to indicate that more desktops could be used, the CPU limitations previously described would have reduced the user experience dramatically. C7 configuration 37

38 Multimedia workload (P40-1B), 72 vgpu users, ESXi 6.5U2, XenDesktop 7.15 MCS We ran the following tests on this workload. CPU usage We populated the GPU-enabled compute host with 72 vgpu-enabled VMs and used the NVIDIA P40-1B profile. With all user VMs powered on and before we started the test, the CPU usage was approximately 11 percent on the GPU-enabled compute host. The following figure shows the performance data for 72 user sessions on the management and GPU-enabled compute hosts. The CPU reached a steady state average of 92.8 percent during the test cycle when all users were logged on to the GPU-enabled compute host. The management host reached a CPU maximum of 6 percent during logoff. GPU utilization We gathered the GPU metrics from the vsphere Client. The GPU usage during the steady state period averaged approximately 39.8 percent and reached a peak usage of 55.3 percent with the multimedia workload. 38 Dell EMC Ready Architectures for VDI

39 Memory For this test run, no memory constraints occurred on the management or GPUenabled compute hosts. Of a total of 768 GB available memory per node, the GPU compute host reached a maximum memory consumption of 633 GB with active memory usage reaching a maximum of 609 GB during the steady-state phase. Each user session consumed 8.7 GB of memory and 8.4 GB of active memory. Memory usage did not vary throughout the test because all vgpu-enabled VM memory was reserved. No memory ballooning or swapping occurred on either host. C7 configuration 39

40 Network usage Network bandwidth was not an issue on this test run with a steady-state peak of approximately 710 Mbps. The busiest period for network traffic was during the logon phase. The compute/gpu hosts reached a peak of 710 Mbps at the start of steady state. IOPS The IOPS graphs and IOPS numbers came from the Nutanix Prism web console. They clearly display the boot storm phase, the initial logon of the desktops, the steady state, and then the logoff phase. The graphs show IOPS data for the cluster and the GPU host. 40 Dell EMC Ready Architectures for VDI

41 The cluster IOPS reached a maximum of 5,791 disk IOPS during the boot storm and averaged 466 disk IOPS during steady state. Based on these numbers, each user session generated 6.4 disk IOPS during steady state. The cluster controller IOPS reached a maximum of 9,361 disk IOPS during the boot storm and averaged 500 disk IOPS during steady state. Based on these numbers, each user session generated 6.9 IOPS in steady state. GPU host disk IOPS The GPU host reached a maximum of 5,771 disk IOPS during the boot storm and averaged 418 disk IOPS during steady state. Based on these numbers, each user session generated 5.8 IOPS in steady state. C7 configuration 41

42 I/O latency The latency graphs and latency numbers came from the Nutanix Prism web console. They clearly display the initial logon of the desktops, the steady state, and then the logoff phase. The graphs show latency data for the cluster and the GPU host. The cluster latency reached a maximum latency of 2.9 ms during the boot storm and averaged 0.37 ms during steady state. The GPU host latency reached a maximum latency of 4.5 ms during the boot storm and averaged 0.35 ms during steady state. 42 Dell EMC Ready Architectures for VDI

43 User experience The following figure shows that the user experience score did not reach the VSI maximum for this test. When we manually interacted with the sessions during steady state, the mouse and window movement were responsive and video playback was good. The baseline performance of 893 indicates that the user experience for this test run was good. The index average reached 1,361, which was well below the threshold of 1,894. Note No disk latency issues occurred during testing. C7 configuration 43

44 Knowledge Worker, 550 users, Hyper-V 2016, XenDesktop 7.15 MCS We ran the following tests on this workload. CPU usage For the Knowledge worker test run, we provisioned 185 VMs on each compute host and 180 on the management host, along with all XenDesktop VM roles. The peak CPU usage was 71 percent on one node at the end of the logon phase, while the steady-state average CPU usage was 60 percent. During steady state, the CVM roles averaged 10 percent on each host while the management roles averaged 0.26 percent of the management host CPU. Consumed memory The memory used during the steady state averaged 516 GB, and the peak memory usage was 539 GB on one node during steady state. 44 Dell EMC Ready Architectures for VDI

45 Network usage The steady-state average network usage was 865 Mbps on the compute hosts. The peak was 3,294 during the boot storm on node compute B. IOPS The peak cluster IOPS for the test run was 7,700 IOPS during boot storm, while the average in steady state was 1,216 IOPS. The peak IOPS for any single node was 4,922 IOPS on the management host during the boot storm. I/O latency The peak I/O latency was 1.4 ms during the boot storm. The average I/O latency during steady state was 1.0 ms. The chart clearly shows a steady low level of I/O latency throughout the test run. C7 configuration 45

46 User experience The baseline performance score of 1,091 indicates that the user experience for this test run was good. The index average reached 1,733, which was well below the threshold of 2,091. A total of nine sessions failed to login or failed to become active. Notes As indicated in the CPU usage definition, the CPU graphs do not consider the extra 35 percent of CPU resources available through the Intel Xenon Gold 6138 processors turbo feature. No disk latency issues occurred during testing. 46 Dell EMC Ready Architectures for VDI

47 Task worker, 1000 users, Hyper-V 2016, XenApp 7.15 MCS (hosted shared) We ran the following tests on this workload. CPU usage For the task worker test run, we provisioned nine VMs on each compute host and eight VMs on the management host along with the XenDesktop/XenApp management roles. The peak CPU usage was 71 percent on one node at the end of the logon phase, while the steady-state average CPU usage was 56 percent. Consumed memory The memory consumed per host during the test run averaged 338 GB during steady state. The peak memory usage was 340 GB on one host during steady state. There was no ballooning or swapping of VM memory during the test run. Each user session consumed 1.02 GB of memory after accounting for CVM and management memory usage. Network usage The average network usage during steady state was 938 Mbps. The peak was 2,157 Mbps during the boot storm phase. C7 configuration 47

48 IOPS The peak cluster IOPS for the test run was 4,001 IOPS during a spike in the steadystate phase. The average during steady state was 439 IOPS. Based on these numbers, each user session generated 0.44 IOPS in steady state. I/O latency The peak I/O latency was 0.8 ms during the boot storm. The average I/O latency during steady state was 0.4 ms. The chart clearly shows a steady low level of I/O latency throughout the test run. 48 Dell EMC Ready Architectures for VDI

49 User experience The baseline performance score of 770 indicates that the user experience for this test run was very good. The index average reached 1,590, which was well below the threshold of 1,770. Notes As indicated in the CPU usage definition, the CPU graphs do not consider the extra 35 percent of CPU resources available through the Intel Xeon Gold 6138 processors turbo feature. C7 configuration 49

50 There was a lack of IOPS during the logon phase in this test. During the steadystate phase, there appeared to be a spike of unknown origin. We assumed that after the logins to the shared hosted desktops were completed, storage I/O destaging occurred from the cache tier to the storage tier of the Nutanix system. This did not appear to affect user experience in a significant way. 50 Dell EMC Ready Architectures for VDI

51 CHAPTER 4 Conclusion Density recommendations...52 Summary...52 Conclusion 51

52 Conclusion Density recommendations We tested all configurations with Microsoft Windows 10 and Microsoft Office Test results provide recommended user densities, as shown in the following table. Table 10 User density recommendations for VMware vsphere ESXi 6.5 U1 with Citrix Virtual Apps and Desktops. Server configuration Workload User density B5 + 3 x P40 Multimedia (virtual PC: P40-2B) 36 C7 Task worker 190 C7 Knowledge worker 160 C7 Power worker 130 C7 + 3 x M60 Power worker (virtual workstation: M60-1Q) 48 C7 + 3 x P40 Multimedia (virtual PC: P40-1B) 72 Table 11 User density recommendations for Microsoft Hyper-V Server configuration Workload User density C7 - Hyper-V 2016 (Shared Desktop) Task worker 342 C7 - Hyper-V 2016 Knowledge worker 185 C7 - Hyper-V 2012 R2 Task worker 210 C7 - Hyper-V 2012 R2 Knowledge worker 185 C7 - Hyper-V 2012 R2 Power worker 145 Summary The configurations for the XC Family devices have been optimized for VDI. We selected the memory and CPU configurations that provide optimal performance. You can change these configurations to meet your own requirements, but keep in mind that changing the memory and CPU configurations from those that have been validated in this document will affect the user density per host. With the introduction of the six-channels-per-cpu requirement for Skylake, the C7 memory configuration recommendation has increased from the previous guidance of 512 GB to 768 GB. This change was necessary to ensure a balanced memory configuration and optimized performance for your VDI solution. The additional memory is advantageous, considering the resulting increase in operating system resource utilization and the enhanced experience for users when they have access to additional memory allocations. 52 Dell EMC Ready Architectures for VDI

53 CHAPTER 5 References This chapter presents the following topics: Dell EMC documentation VMware documentation Citrix resources...54 References 53

54 References Dell EMC documentation VMware documentation Citrix resources The following Dell EMC documentation provides additional and relevant information. Access to these documents depends on your login credentials. If you do not have access to a document, contact your Dell EMC representative. Also see the Dell EMC VDI Information Hub for a more complete list of VDI resources. Dell EMC Virtual Desktop Infrastructure Dell EMC XC Series and XC Core Technical Resource Center This document is part of the documentation set for this architecture, which includes the following: Dell EMC Ready Architectures for VDI: Designs for Citrix Virtual Apps and Desktops on XC Family Design Guide Dell EMC Ready Architectures for VDI: Designs for Citrix Virtual Apps and Desktops on XC Family Deployment Guide Dell EMC Ready Architectures for VDI: Designs for Citrix Virtual Apps and Desktops on XC Family Validation Guide The following VMware resources provide additional and relevant information: VMware vsphere documentation VMware Compatibility Guide The following Citrix resources provide additional and relevant information: XenDesktop and XenApp 7.15 LTSR: System Requirements Citrix VDI Handbook and Best Practices Citrix deployment guides Citrix StoreFront Proof of Concept Implementation Guide Install and Configure 54 Dell EMC Ready Architectures for VDI