Dell EMC Ready Architectures for VDI

Transcription

1 Dell EMC Ready Architectures for VDI Designs for VMware Horizon on VxRail and VSAN Ready Nodes September 2018 H Validation Guide Abstract This validation guide describes the architecture and performance of the integration of VMware Horizon components for virtual desktop infrastructure (VDI) on Dell EMC VSAN Ready Nodes and Dell EMC VxRail with 14th generation Dell EMC PowerEdge Servers. Dell EMC Solutions

2 Copyright 2018 Dell Inc. or its subsidiaries. All rights reserved. Published September 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 Validated software resources... 9 Validated system version matrix... 9 Virtual networking configuration... 9 Management server infrastructure...10 NVIDIA GRID License Server...10 SQL Server databases DNS High availability...11 VMware Horizon architecture Chapter 3 Solution Performance and Testing 13 Testing process...14 Resource monitoring Load generation Profiles and workloads Desktop VM test configurations Test results and analysis B5 configuration...19 C7 configuration Chapter 4 Conclusion 57 Test results and density recommendations...58 Conclusions Chapter 5 References 59 Dell EMC documentation...60 VMware documentation NVIDIA documentation 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 VxRail appliances and vsan Ready Nodes with VMware Horizon 7. 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 VMware Horizon 7 VDI components on VxRail appliances or vsan Ready Nodes. 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... 9 Management server infrastructure High availability VMware Horizon architecture...12 Test Environment Configuration and Best Practices 7

8 Test Environment Configuration and Best Practices Validated hardware resources Dell EMC validated the solutions with the specific hardware resources listed in this section. Enterprise platforms We used the VxRail appliances and vsan Ready Nodes hardware components listed in the following table. For reference, we have designated the configurations as B5 and C7, which are referenced in the test results sections. Table 1 Validated hardware configurations Config Enterprise platform CPU Memory RAID ctlr BOSS HD config Network B5 V570F R740XD 5120 Gold (14- core 2.2 GHz) 384 2,400 MT/s HBA x 120/240G B M.2 2 x 400 GB SSD (Cache) 4 x 1.92 TB HDD (Capacity) 4 x Intel X710 rndc C7 V570F R740XD 6138 Gold (20- core 2.0 GHz) 768 2,667 MT/s HBA x 120/2,40G B M.2 2 x 800 GB SSD (Cache) 6 x 1.92 TB HDD (Capacity) 4 x Intel X710 rndc 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 NVIDIA Maxwell GPUs and a total of 32 GB GDDR5 memory per card (8 GB per GPU) NVIDIA Pascal P40 A dual-slot 10.5-inch PCI Express Gen3 graphics card featuring a single high-end NVIDIA Pascal GPU. With a total of 24 GB GDDR5 memory per card NVIDIA Pascal P4 A single-slot 10.5-inch PCI Express Gen3 graphics card featuring a single high-end NVIDIA Pascal GPU with a total of 8 GB GDDR5 memory per card Network hardware The following network hardware was used 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 performance 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 performance 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 listed in the following table. Component Hypervisor Description/Version ESXi 6.5 (VxRail), ESXi 6.7 (VSRN) Broker technology VMware Horizon 7 version Broker database Microsoft SQL Server 2016 Management VM OS Virtual desktop OS Microsoft Windows Server 2016 (Connection Server & DB) Microsoft Windows 10 Enterprise Office application suite Microsoft Office Professional 2016 Login VSI test suite Version 4.1 Platform VxRail v 4.5 or VSRN with vsan v 6.5 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 2 Version matrix for tested system Server Configurat ion Hypervisor Hypervisor Version Hypervisor Build Bios VxRail version Windows 10 Version Nvidia vgpu Version B5 ESXi 6.5 U2b n/a B5 + 2 x M10 ESXi 6.5 Patch B5 + 6 x P4 ESXi n/a C7 ESXi 6.5 U2b n/a C7 + 3 x P40 ESXi 6.5 Patch Virtual networking configuration The network configuration for the uses a 10 GbE converged infrastructure model. All required VLANs traverse two 10 GbE NICs configured in an active/active team. We used the following VLAN configurations for the compute and management hosts in our validation testing. VLAN configuration: Validated software resources 9

10 Test Environment Configuration and Best Practices Management VLAN: Configured for hypervisor infrastructure traffic L3 routed via core switch VDI VLAN: Configured for VDI session traffic L3 routed via core switch VMware vsan VLAN: Configured for VMware vsan traffic L2 switched only via ToR switch vmotion VLAN: Configured for Live Migration traffic L2 switched only, trunked from Core (HA only) VDI Management VLAN: Configured for VDI infrastructure traffic L3 routed via core switch A VLAN idrac is configured for all hardware management traffic L3 routed via core switch Management server infrastructure The management server component sizing recommendations are listed in the following table. Table 3 Sizing for VxRail appliance, RDSH, and NVIDIA GRID license server (optional) Component vcpu RAM (GB) NIC OS + data vdisk (GB) Tier 2 volume (GB) VMware vcenter Appliance Platform Services Controller Horizon Connection Server SQL Server (VMDK) File server (VMDK) VxRail Appliance Manager Log Insight RDSH VM NVIDIA GRID License Server NVIDIA GRID License Server 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 10 Dell EMC Ready Architectures for VDI

11 Test Environment Configuration and Best Practices SQL Server databases 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 Horizon Connection 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. SQL Server databases 11

12 Test Environment Configuration and Best Practices VMware Horizon architecture The following figure shows the Horizon communication flow. Figure 1 VMware Horizon 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 4 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 5 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 5 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 Linked vs. Instant Clones Horizon supports two provisioning methods that deliver space-optimized virtual desktop pools: Linked Clones and Instant Clones. The user density per host is not impacted by using one over the other. The differences in the test graphs for these two methods are a result of the following processes: For Linked Clones, all the VMs are rebooted before the test starts to simulate a boot storm. 16 Dell EMC Ready Architectures for VDI

17 For Instant Clones, the VMs are rebooted after the session logs off, because when a user logs out of the instant clone, the clone is destroyed and recreated for the next user. The following figure shows the differences. Desktop VM test configurations The following table summarizes the compute VM configurations for the profiles and workloads that we tested. Table 6 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. Desktop VM test configurations 17

18 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 Validated system version matrix 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 7 Test results summary Server configuration Workload User density Protocol Avg CPU Avg active memory Avg IOPS / user B5 Task worker 120 PCoIP 88% 97 GB 4.2 B5 Knowledge worker 105 PCoIP 87% 117 GB 5.6 B5 Power worker 80 PCoIP 87% 103 GB 3.0 B5 + 2 x M10 Power worker (virtual PC: M10-1B) 64 Blast 48% 269 GB 12 B5 + 6 x P4 Power worker (virtual PC: P4-1B) 48 Blast 60% 288 GB 28 B5 + 6 x P4 Multimedia (virtual workstation: P4-2Q) 24 Blast 61% 144 GB 7 C7 Task worker 165 PCoIP 89% 116 GB 4 C7 Knowledge worker 140 PCoIP 89% 127 GB 5 C7 Power worker 115 PCoIP 86% 130 GB 5 C7 + 3 x P40 Multimedia (Virtual PC: P40-1B) 72 Blast 95% 577 GB 57 C7 + 3 x P40 Multimedia (Virtual workstation: P40-2Q) 36 Blast 59% 288 GB 60 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 18 Dell EMC Ready Architectures for VDI

19 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. B5 configuration We performed the following task worker, knowledge worker, power worker, and multimedia performance testing on the vsan Ready Nodes-R740 in the B5 configuration described in Validated hardware resources on page 8. These results also apply to the VxRail B5 configuration, as the configurations are the same. The ESXi versions were different at the time of release of this document, but based on previous testing, moving from 6.5 to 6.7 did not affect user density results. Task Worker, 350 users, ESXi 6.x, Horizon 7 v7.x We ran the following tests on this workload. CPU usage We populated each host with 120 virtual machines per compute host and 110 virtual machines on the compute/management host in addition to the management VMs. With all user VMs powered on and before we started the test, the CPU usage was approximately 23 percent on the compute hosts. The following figure shows the performance data for 120 user sessions per compute host. The CPU reached a steady-state average of 86 percent across the two compute hosts during the test cycle when 120 users were logged on to each compute host. The average CPU usage for the combined compute/management host was 88 percent during steady state. B5 configuration 19

20 Memory With regard to active memory consumption for the cluster, out of a total of 384 GB available memory per node, no issues occurred. The compute hosts reached a maximum active memory consumption of 93 GB. No ballooning or memory swapping occurred during any point during testing. Network usage Network bandwidth was not an issue on this test run with a steady-state peak of approximately 813 MBps on the compute/management host and a peak of 764 Mbps on one off the compute hosts. The busiest period for network traffic was during logoff after testing had completed. One of the hosts reached a peak of 5,888 MBps during the re-creation of the instant clones. IOPS The IOPS graphs clearly show the initial logon of the desktops, the steady-state and logoff phases, and finally the re-creation of the desktops after testing was complete. The graph displays the disk IOPS figure for the VSAN cluster. 20 Dell EMC Ready Architectures for VDI

21 The cluster reached a maximum total (read + write) of 7,286 disk IOPS during the instant clone re-creation period after testing and a total average of 1,479 IOPS during steady state. Disk I/O latency Disk I/O latency was not an issue during the Login VSI testing period of this test run. Total latency (read + write) reached its maximum of approximately 4,430 ms during the boot storm. The average total latency during steady state was 1,523 ms. User experience During this test, the user experience score did not reach the Login VSI maximum for this test, as shown in the following figure. B5 configuration 21

22 Knowledge Worker, 305 users, ESXi 6.x, Horizon 7.x We ran the following tests for this workload. CPU usage We populated the compute hosts with 105 VMs per host and the combined management/compute host with 90 VMs in addition to the management VMs. With all user virtual machines powered on and before we started the test, the CPU usage on the compute hosts was approximately 18 percent. The following figure shows the performance data for 105 user sessions per compute host. The CPU reached a steady state average of 87 percent across the two compute hosts when 105 users were logged on to each host. The average CPU usage on the combined compute/management host was 86 percent during steady state. Memory With regard to memory consumption for the cluster, out of a total of 384 GB available memory per node, no memory resource issues occurred. Active memory usage on the compute hosts reached a maximum of 117 GB. No ballooning or memory swapping occurred at any point during testing. 22 Dell EMC Ready Architectures for VDI

23 Network usage Network bandwidth was not an issue on this test run with a steady-state peak of approximately 1,240 MBps on the combined compute/management host. One of the compute hosts reached a maximum of 967 Mbps during steady state. The busiest period for network traffic was during logoff after testing was complete. One of the hosts reached a peak of 5,896 Mbps during re-creation of the instant clones. IOPS The IOPS graphs clearly display the initial logon of the desktops, the steady state and logoff phases, and finally the re-creation of the desktops after testing was complete. The graph shows the disk IOPS figure for the VSAN cluster. The cluster reached a maximum total (read + write) of 7,244 disk IOPS during the instant clone re-creation period after testing. The total average during steady state was 1,731 disk IOPS. B5 configuration 23

24 Disk I/O latency Disk I/O latency was not an issue during the Login VSI testing period of this test run.total latency (read + write) reached its maximum of approximately 6,380 ms during the boot storm. The average total latency during steady state was 1,766 ms. User experience The following figure shows that during this test, the user experience score did not reach the Login VSI maximum for this test. 24 Dell EMC Ready Architectures for VDI

25 Power Worker, 230 users, ESXi 6.x, Horizon 7 v7.x We ran the following tests on this workload. CPU usage We populated the compute hosts with 80 virtual machines per host and the combined compute/management host with 70 virtual machines in addition to the management VMs. With all user VMs powered on and before we started the test, the CPU usage was approximately 11 percent on the compute hosts before reboot. The following figure shows that the CPU reached a steady state average of 87 percent across the two compute hosts when 80 users are logged on to each host. The average CPU usage on the combined compute/management host was 87 percent during steady state. Memory With regard to memory consumption for the cluster, with a total of 384 GB memory per node, no memory resource issues occurred. The compute hosts reached a maximum active memory consumption of 97 GB. No ballooning or memory swapping occurred at any point during testing. B5 configuration 25

26 Network usage Network bandwidth was not an issue on this test run with a steady-state peak of approximately 2,206 Mbps on the combined compute/management host and 1,569 Mbps on one of the compute hosts. The busiest period for network traffic was during the logoff of the pool after testing was complete. The combined hosts total usage reached a peak of 4,318 Mbps during re-creation of the instant clones. IOPS The IOPS graphs clearly display the initial logon of the desktops, the steady state and logoff phases, and finally the re-creation of the desktops after testing was complete. The graph shows the disk IOPS figure for the VSAN cluster. The cluster reached a maximum of 5,652 total (read + write) disk IOPS during the instant clone re-creation period after testing. The average total disk IOPS was 710 during steady state. 26 Dell EMC Ready Architectures for VDI

27 Disk I/O latency Disk I/O latency was not an issue during the Login VSI testing period of this test run. The total latency (read + write) reached a maximum of approximately 2,581 ms during steady state. The average total latency during steady state was 778 ms. User experience During the test, the user experience score did not reach the Login VSI maximum for this test, as shown in the following figure. B5 configuration 27

28 Power Worker (M10-1B), 64 vgpu users, ESXi 6.x, Horizon 7.x We ran the following tests for this workload. CPU usage WE populated the compute host with two NVIDIA M10 cards. We ran the LVSI power worker workload for this test. CPU usage was well below the 95 percent mark using 64 clients with the M10-1B profile. Memory Because memory usage is reserved with vgpu applications, consumed and active memory showed a value slightly above the reserved value of 256 GB (64 clients x 4 GB per client). Installed memory for this test was 512 GB. During tests no memory ballooning or swapping occurred. 28 Dell EMC Ready Architectures for VDI

29 Network usage Network bandwidth usage was consistent at about 280 Mbits/sec during steady state. Network usage was not an issue during tests. GPU usage Average GPU processor usage was in the range of 60 percent during steady state. This utilization is typical for the power worker load, because it is not considered a high load in GPU terms. B5 configuration 29

30 Multimedia (P4-1B), 24 Users, ESXi 6.x, Horizon 7 We ran the following tests for this workload. CPU usage For the multimedia test run, we provisioned 24 1B vgpu VMs on compute host 3, which had six P4 graphics cards three cards in 16 PCIe slots and three cards in 8 PCIe slots. Steady state average was 60 percent. Memory The memory average active during the test was 288 GB. Since this was vgpu testing, all memory was reserved. 30 Dell EMC Ready Architectures for VDI

31 Network usage Network bandwidth was not an issue on this test run with a steady state peak of approximately 37 MBps and a logon spike of 106 MBps. IOPS The peak VSAN IOPS for the test run was 642 write and 449 read IOPS during the steady state, while the average in steady state was 63 read and 280 write IOPS. The average IOPS per user for this test was 7.1. B5 configuration 31

32 I/O Latency The peak I/O read latency was ms and IO write was latency ms. The average read latency during steady state was ms, while the average write latency was ms. The chart clearly shows a very low and steady level of IO latency throughout the test run. GPU usage The GPU metrics came form the vsphere client. The GPU usage during the steady state period averaged approximately percent and reached a peak usage of 20 percent with the multimedia workload. 32 Dell EMC Ready Architectures for VDI

33 User experience The baseline performance of 953 indicated that the user experience for this test run was good. The Index average reached 1,194 well below the threshold of 1,953. Multimedia (P4-2Q), 24 Users, ESXi 6.x, Horizon 7.x We ran the following tests on this workload. CPU usage For the multimedia test run, we provisioned 24 1B vgpu VMs on compute host 3, which had six P4 Graphics cards three cards in 16 PCIe slots and three cards in 8 PCIe slots. Steady state average was 60 percent. B5 configuration 33

34 Memory The memory average active during the test was 144 GB. Since this was vgpu testing, all memory was reserved. Network usage Network bandwidth was not an issue on this test run with a steady state peak of approximately 96 MBps and a logon spike of 141 MBps. 34 Dell EMC Ready Architectures for VDI

35 IOPS The peak VSAN IOPS for the test run was 722 write and 343 read IOPS during the steady state, while the average during steady state was 192 read and 479 write IOPS. The average IOPS per user for this test was I/O latency The peak I/O read latency was ms and I/O write latency was ms. The average read latency during steady state was ms, while the average write latency was ms. The chart clearly shows a very low and steady level of IO latency throughout the test run. B5 configuration 35

36 GPU usage The GPU metrics came form the vsphere client. The GPU usage during the steady state period averaged approximately percent and reached a peak usage of 26 percent with the multimedia workload. User experience The baseline performance of 953 indicated that the user experience for this test run was good. The Index average reached 1,194, which was well below the threshold of 1, Dell EMC Ready Architectures for VDI

37 B5 configuration 37

38 C7 configuration We performed the following multimedia performance testing on the vsan Ready Node in the C7 configuration described in Validated hardware resources on page 8. These results also apply to the VxRail C7 configuration, as the configurations are the same. The ESXi versions were different at the time of release of this document, but based on previous testing, moving from 6.5 to 6.7 did not affect user density results. Task Worker, 475 users, ESXi 6.x, Horizon 7 v7.x We ran the following tests for this workload. CPU usage We populated each compute host with 165 virtual machines per compute host and the combined compute/management host with 145 user VMs in addition to the management VMs. With all user VMs powered on and before starting the test, the CPU usage was approximately 19 percent on the compute hosts. The following figure shows the performance data for 165 user sessions per compute host. The CPU reached a steady state average of 88 percent across the two compute hosts when 165 users were logged on to each host. The combined compute/ management host average CPU usage was 79 percent during steady state. Memory With regard to memory consumption for the cluster, with a total of 768 GB of memory per node, no issues occurred with memory resources. The compute hosts reached a maximum memory consumption of 385 GB with active memory usage reaching a maximum of 291 GB. No ballooning or memory swapping occurred at any point during testing. 38 Dell EMC Ready Architectures for VDI

39 Network usage Network bandwidth was not an issue on this test run with a steady state peak of approximately 780 Mbps on the combined compute/management host and a peak of 639 Mbps on one of the compute only hosts. The busiest period for network traffic was during the recreation of the instant clones after testing had completed. The hosts reached a combined peak of 3,274 Mbps during the deletion and recreation of the instant clones. IOPS The IOPS graphs clearly display the initial logon of the desktops, the steady state and logoff phases, and finally the recreation of the desktops after testing was complete. The graph displays the disk IOPS figure for the VSAN cluster. C7 configuration 39

40 The cluster reached a maximum total (read + write) of 8,053 disk IOPS during the instant clone recreation period after testing and the steady state total average was 1,811 disk IOPS. Disk I/O latency Disk I/O latency was not an issue during the Login VSI testing period of this test run. Total latency (read + write) reached its maximum of approximately 2.2 ms during the recreation of the instant clones. This was well below the 20 ms threshold where latency becomes potentially troublesome. The average total latency during steady state was 1.4 ms. User experience During the test, the user experience score did not reach the Login VSI maximum for this test, as shown in the following figure. 40 Dell EMC Ready Architectures for VDI

41 Knowledge Worker, 410 users, ESXi 6.x, Horizon 7 v7.x We ran the following tests for this workload. CPU usage We populated each compute host with 140 virtual machines per host and the combined compute/management host with 130 user VMs in addition to the management VMs. With all user VMs powered on and before starting the test, the CPU usage was approximately 16 percent on the compute hosts. The following figure shows the performance data for 140 user sessions per compute host. The CPU reached a steady state average of 89 percent across the two compute hosts when 140 users were logged on to each host. The combined compute/ management host average CPU usage was 82 percent during steady state. C7 configuration 41

42 Memory With regard to memory consumption for the cluster, with a total of 768 GB of memory per node, no issues occurred with memory resources. The compute hosts reached a maximum memory consumption of 452 GB, while active memory usage reached a maximum of 373 GB. No ballooning or memory swapping occurred at any point during testing. 5in 42 Dell EMC Ready Architectures for VDI

43 Network usage Network bandwidth was not an issue on this test run with a steady state peak of approximately 1,636 Mbps on the combined compute/management host and a peak of 1,157 Mbps on one of the compute hosts. The busiest period for network traffic was during the recreation of the instant clones after testing was complete. One of the hosts reached a peak of 4,133 Mbps during the logoff period and 3,133 Mbps during the recreation of the instant clones. IOPS The IOPS graphs clearly display the initial logon of the desktops, the steady state and logoff phases, and finally the recreation of the desktops after testing was complete. The graph displays the disk IOPS figure for the VSAN cluster. The cluster reached a maximum total (read + write) of disk 7,172 IOPS during the instant clone recreation period after testing. The steady state total average was 1,948 disk IOPS. C7 configuration 43

44 Disk I/O latency Disk I/O latency was not an issue during the Login VSI testing period of this test run. Total latency (read + write) reached its maximum of approximately 2.8 ms during the recreation of the instant clones. This was well below the 20 ms threshold where latency becomes potentially troublesome. The average total latency during steady state was 1.9 ms. User experience The following figure shows that the user experience score did not reach the Login VSI maximum for this test. 44 Dell EMC Ready Architectures for VDI

45 Power Worker, 305 users, ESXi 6.x, Horizon 7 v7.x We ran the following tests for this workload. CPU We populated each compute host with 115 virtual machines per host and the combined compute/management host with 105 user VMs in addition to the management VMs. With all user VMs powered on and before starting the test, the CPU usage was approximately 13 percent on the compute hosts. The following figure shows the performance data for 115 user sessions per compute host. The CPU reached a steady state average of 86 percent across the two compute hosts when 115 users were logged on to each host. The combined compute/ management host average CPU usage was 81 percent during steady state. C7 configuration 45

46 Memory With regard to memory consumption for the cluster, with a total of 768 GB of memory per node, no issues occurred with memory resources. The compute hosts reached a maximum memory consumption of 504 GB with active memory usage reaching a maximum of 414 GB. No ballooning or memory swapping occurred at any point during testing. 46 Dell EMC Ready Architectures for VDI

47 Network usage Network bandwidth was not an issue on this test run with a steady state peak of approximately 1,817 Mbps on the combined compute/management host and a peak of 1,178 Mbps on one of the compute hosts. The busiest period for network traffic was during the recreation of the instant clones after testing was complete. The combined hosts total usage reached a peak of 4,659 Mbps during the logon phase and 3,003 Mbps during the deletion and recreation of the instant clones. IOPS The IOPS graphs clearly display the initial logon of the desktops, the steady state and logoff phases, and finally the recreation of the desktops after testing was complete. The graph displays the disk IOPS figure for the VSAN cluster. The cluster reached a maximum total (read + write) of 4,604 disk IOPS during the instant clone recreation period after testing. The steady state total average was 1,662 disk IOPS. C7 configuration 47

48 Disk I/O latency Disk I/O latency was not an issue during the Login VSI testing period of this test run. Total latency (read + write) reached its maximum at approximately 2.6 ms during the logon phase. This was well below the 20 ms threshold where latency becomes potentially troublesome. The average total latency during steady state was 1.9 ms. User experience The following figure shows that the user experience score did not reach the Login VSI maximum for this test. 48 Dell EMC Ready Architectures for VDI

49 Multimedia (P40-1B), 72 users, ESXi 6.5, Horizon 7.x We ran the following tests for this workload. CPU usage We populated the compute host with 72 vgpu-enabled virtual machines with 1 GB of framebuffer, all placed on the host with three P40 GPU cards installed. With all user virtual machines powered on and before starting the test, the CPU usage was approximately 10 percent. The following figure shows the performance data for 72 user sessions per host. The CPU reached a steady state average of 95 percent on the GPU-enabled compute host during the test cycle when 72 users were logged on to the host. C7 configuration 49

50 GPU usage We populated the compute host server with three NVIDIA P40 GPU cards. The following figure shows the performance of the three GPU cards through the test period. During the steady state period the GPUs had an average usage of approximately 36 percent, which indicates that even with the maximum population of 72 vgpu VMs, the multimedia workload did not drive the GPUs close to their maximum. The GPU reached a peak usage of 41 percent. Memory With regard to memory consumption for the GPU-enabled host, out of a total of 768 GB of available memory, there was no constraint on resources. The host reached a maximum memory consumption of 636 GB with active memory usage reaching a max 50 Dell EMC Ready Architectures for VDI

51 of 577 GB. There was no memory swapping or ballooning on the host. The graphs display a flat line for both metrics as all assigned memory (8 GB) was reserved for each vgpu-enabled VM. Network usage Network bandwidth was not an issue on this test run with a steady state peak of approximately 1,950 Mbps. This was also the busiest period for network traffic during the test run. IOPS The IOPS graphs and IOPS numbers came from the vcenter web console. The graphs clearly display the initial boot storm, the logon of the desktops, the steady state period, and the logoff phase. The graph displays the disk IOPS figure for the VSAN cluster. The cluster reached a maximum of 6,814 (6,373 read/441 write) disk IOPS during the reboot of all the VMs before starting testing and 4,097 (2,513 read/1,584 write) IOPS at the start of steady state. C7 configuration 51

52 Disk I/O latency Disk I/O latency was not an issue during the Login VSI testing period of this test run. The maximum latency reached was approximately 1.6 ms at the beginning of steady state and 2.5 ms during the reboot of all the VMs before starting test. This was well below the 20 ms threshold where latency becomes potentially troublesome. User experience During this test, the user experience score did not reach the Login VSI maximum for this test, as shown in the following figure, indicating that the user density we tested did not place any strain on system resources. We logged into a session under test using a Wyse 3040 Thin Client to assess the user experience while the system was running the steady state phase of test. The user experience was deemed to be excellent with the graphics intensive Google Earth and Bouncing Objects performing smoothly with no hanging or jerking movement. 52 Dell EMC Ready Architectures for VDI

53 Note The R740xd "C7" configuration for the P40-2Q testing is not optimal, as the testing has shown. For a three-p40 host configuration using the P40-2Q, we recommend the R740xd "B5" configuration. Multimedia (P40-2Q), 36 users, ESXi 6.5, Horizon 7.x We ran the following tests for this workload. CPU usage We populated the compute host with 36 vgpu-enabled virtual machines with 2 GB of framebuffer, all placed on the host with three P40 GPU cards installed. With all user virtual machines powered on and before starting the test, the CPU usage was approximately 5 percent. The following figure shows the performance data for 36 user sessions per host. The CPU reached a steady state average of 59 percent during the test cycle when all 36 users were logged on to the host. GPU usage We populated the compute host server with three NVIDIA P40 GPU cards. The following figure shows the performance of the three GPU cards through the test period. During the steady state period the GPUs had an average usage of approximately 32 percent, which indicates that, even with the maximum population of 36 2Q profile vgpu VMs, the multimedia workload did not drive the GPUs close to their maximum. The GPU reached a peak usage of 38 percent. C7 configuration 53

54 Memory With regard to memory consumption for the GPU enabled host, out of a total of 768 GB of available memory, there was no constraint on resources. The host reached a maximum memory consumption of 343 GB with active memory usage reaching a max of 288 GB. There was no memory swapping or ballooning on the host. The graphs display a flat line for both metrics as all assigned memory (8 GB) was reserved for each vgpu enabled VM. Note The testing completed here with the C7 configuration with 768 Gb of memory is not an optimal configuration for this workload with this graphics user profile. In this case we would recommend using the B5 configuration. The platform configurations that we have validated can be modified to suit your needs, but note that changing the memory and CPU will have an impact on the user density of the host. 54 Dell EMC Ready Architectures for VDI

55 Network usage Network bandwidth was not an issue on this test run with a steady state peak of approximately 1,366 Mbps just after the login phase had completed. This was also the busiest period for network traffic during the test run. The management host reached a peak of 241 Mbps, also during the steady state phase. IOPS The IOPS graphs and IOPS numbers came from the vcenter web console. The graphs clearly display the initial boot storm, the logon of the desktops, the steady state period, and the logoff phase. The graph displays the disk IOPS figure for the VSAN cluster. The cluster reached a maximum of 3,097 disk IOPS (2,120 read and 977 write) just before the start of steady state. Disk I/O Latency Disk I/O latency was not an issue during the Login VSI testing period of this test run. The maximum latency reached was approximately 13 ms during the reboot of all desktops before test start and 1.3 ms at the start of steady state. This was well below the 20 ms threshold where latency becomes potentially troublesome. C7 configuration 55

56 User experience During this test, the user experience score did not reach the Login VSI maximum for this test, as shown in the following figure, indicating that the user density we tested did not place any strain on system resources. We logged into a session under test using a Wyse 3040 Thin Client to assess the user experience while the system was running the steady state phase of test. The user experience was deemed to be excellent with the graphics intensive Google Earth and Bouncing Objects performing smoothly with no hanging or jerking movement. 56 Dell EMC Ready Architectures for VDI

57 CHAPTER 4 Conclusion Test results and density recommendations Conclusions...58 Conclusion 57

58 Conclusion Test results and density recommendations Test results provided recommended user densities. The user densities in the following table were achieved by following the VMware best practices of FTT = 1 and a reserved slack space of 30 percent. All configurations were tested with Microsoft Windows 10 and Microsoft Office Table 8 User density recommendations for VMware vsphere ESXi 6.5 or 6.7 with VMware Horizon Server configuration Workload User density B5 Task worker 120 B5 Knowledge worker 105 B5 Power worker 80 B5 + 2x M10 Power worker (Virtual PC: M10-1B) 64 B5 + 6x P4 Power worker (Virtual PC: P4-1B) 48 B5 + 6x P4 Multimedia (Virtual workstation: P4-2Q) 24 C7 Task worker 165 C7 Knowledge worker 140 C7 Power worker 115 C7 + 3x P40 Multimedia (Virtual PC: P40-1B) 72 C7 + 3x P40 Multimedia (Virtual workstation: P40-2Q) 36 Conclusions Both vsan Ready Nodes and VxRail appliances reach similar user densities when configured similarly, as demonstrated during this validation. Remember to use VxRail sizing tools to reserve resources for management tools when designing your VDI infrastructure. The configurations for the VxRail appliances and vsan Ready Nodes 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. 58 Dell EMC Ready Architectures for VDI