Technical white paper HP Client Virtualization SMB Reference Architecture for Windows Server 212 Affordable, high-performing desktop virtualization on HP ProLiant DL38p Gen8 and Windows Server 212 Table of contents Introduction 2 A high-performing, cost-effective solution 2 The HP and Microsoft VDI solution 2 Server hardware: HP ProLiant DL38p Gen8 3 Endpoint devices: HP Thin Clients 3 Solution software: Windows Server 212 4 Fully tested, ready to run 4 Performance testing software 4 Initial scalability tests 5 Recommended reference architecture tests 9 Reference architecture bill of materials 11 Conclusion 12 For more information 12 Call to action 12
Introduction Client virtualization used to be considered a solution best suited for large enterprises, but with improvements in virtualization software, the growing popularity of BYOD (bring your own device) plans, and lower overall costs per user, client virtualization has entered the mainstream small and medium-sized business (SMB) market. Many SMBs are now looking at client virtualization as a way to meet their employees demands for flexibility while retaining control of the computing infrastructure for IT, and are excited about the increased security and efficiency, as well as the potential cost savings client virtualization can deliver. With a Virtual Desktop Infrastructure (VDI) client virtualization solution, users desktops run in server-based virtual machines, making them available remotely over a network. These centralized desktops offer a number of distinct benefits for SMBs: Flexible desktop delivery: VDI makes flexible work scenarios for employees possible, including working from home, hot desking (multiple workers using a single physical work station or surface during different time periods), and supporting contractor or offshore information workers. Increased data security and compliance: A VDI solution keeps data safe in a central location and helps eliminate the risk of laptop data theft. In addition, centralized tracking helps simplify the burden of regulatory compliance. Easy and efficient management: IT can install desktop operating systems and applications once, instead of having to install them on every desktop in the organization. This makes it easier to provision and refresh desktops and applications rapidly and on demand. Hewlett-Packard (HP) and Microsoft worked together to design a VDI solution that is ideal for SMBs, as well as regional or divisional offices of large organizations. Based on the HP ProLiant DL38p Gen8 server, the HP Client Virtualization SMB Reference Architecture for Windows Server 212 (HP CV SMB RA for Windows Server 212) makes it easy to deploy a high-performing, affordable virtual desktop solution. And because the design of the HP CV SMB RA for Windows Server 212 leverages the knowledge gained during extensive performance and scalability tests, customers can deploy the solution with confidence. A high-performing, cost-effective solution The HP CV SMB RA for Windows Server 212 makes it simple to deploy a high-performance virtual desktop solution without a high price tag. The HP CV SMB for Windows Server 212 offers several key advantages: Provides leading performance Performance testing, detailed later in this document, confirms that the HP CV SMB RA for Windows Server 212 can support at least 175 virtual desktops per host server with excellent response time. In addition, customers can benefit from performance advancements with HP ProLiant Gen8 servers, ensuring that the solution can comfortably meet the needs of their organizations. Makes deployment easy It is easy to set up the HP CV SMB RA for Windows Server 212 solution on the HP ProLiant server and configure the desktop images, so virtual desktops can be quickly delivered to users. Because all of the software comes from Microsoft, there is no need to work with a third-party vendor, and licensing costs are kept to a minimum. Customers can start with the sample bill of materials included in the reference architecture, and then tailor the solution to meet their unique needs. Includes everything needed With the HP CV SMB RA for Windows Server 212, there is no additional infrastructure to deploy no need for additional management servers, databases, or storage area networks (SANs), and no need for additional virtualization or management software. Everything needed to run the VDI VMs is included in the solution. The HP and Microsoft VDI solution The HP CV SMB RA for Windows Server 212 is based on the HP ProLiant DL38p Gen8 Servers, HP Thin Clients, and Windows Server 212 from Microsoft. The sections that follow describe the hardware and software components of the solution in more detail. 2
Server hardware: HP ProLiant DL38p Gen8 The HP ProLiant DL38p Gen8 server provides a full solution that is modular and scalable for an array of customer needs. For the HP CV SMB RA for Windows Server 212, the HP ProLiant DL38p Gen8 server can be configured to support several user configurations running a standard medium user workload. A medium user workload represents a knowledge worker completing typical desktop activities such as using email, running productivity applications, and viewing high-definition (HD) video. The HP ProLiant DL38p Gen8 offers Intel Xeon E5-26 series processors, up to 768 GB of memory, PCIe expansion slots, two 1 Gigabit network ports, eight small form factor (SFF) drives with an embedded HP Smart Array controller as well as industry leading HP Integrated Lights-Out 4 (ilo 4) management capabilities. Endpoint devices: HP Thin Clients The HP and Microsoft VDI solution can provide virtual desktops to almost any device form factor: thin clients, desktop PCs, laptops, tablets, and smartphones. The HP CV SMB RA for Windows Server 212 was tested with the following recommended HP thin clients. Flexible Series Thin Clients: HP t61 or HP t51 Flexible Series Windows Embedded Standard 7 (WES7) Thin Clients are ideal for customers requiring an advanced graphical user experience. Customers selecting the HP t61 over the HP t51 will benefit from a more powerful CPU, graphics hardware acceleration, quad-display, internal dual-antenna Wi-Fi, Fiber NIC, and PCIe expansion bay options. Customers who purchase the HP t61 typically have a greater mix of remotely virtualized apps and locally embedded apps. Both the HP t61 family and the HP t51 come standard with dual core CPUs and legacy ports for powerful, flexible connectivity to a broad array of peripheral devices. Smart Zero Technology can also be deployed on the Flexible Series Thin Clients, providing the zero touch experience at the endpoint device. For more information, see HP Flexible Series Thin Clients. Smart Zero Clients: HP Smart Zero Clients, like the Power Over Ethernet, All-in-One HP t41, offer customers great value by combining zero client manageability with cost-effective hardware and performance. Leveraging HP s Smart Zero technology, HP Smart Zero Clients intelligently boot and connect to the HP and Microsoft VDI solution without user interaction. Upon verification and confirmation of user credentials, users can begin their normal activities. The Smart Zero technology supports multiple protocols and can be reprogrammed on the fly, delivering a nocompromise, intelligent zero client experience for remote and cloud computing environments. With HP Smart Zero technology installed on the virtualization infrastructure, users can be up and running in seconds with no configuration or management required on the device side. Just set up your server, boot the client, and connect. HP Smart Zero technology combines the benefits of a zero client with HP auto-sensing technology that automatically connects to the network and searches for the right client virtualization infrastructure and downloads everything it needs to deliver a robust user experience. The user is up and running quickly with no local user interface and just three steps, from log on to productivity. 3
For more information, see HP Smart Zero Clients and Server-Side Smart Zero Technology Component. Solution software: Windows Server 212 The HP and Microsoft VDI solution is powered by Windows Server 212, and uses Microsoft s Remote Desktop Protocol (RDP) to provide the remote desktop or remote application to the user. Microsoft RemoteFX enables the delivery of a robust Windows user experience across a range of scenarios. In Windows Server 212, enhancements provide a richer and more seamless experience on all types of networks and devices. Windows Server 212 is an ideal choice for desktop virtualization. It includes everything needed to set up a VDI solution; the connection broker, web access, gateway, and other necessary components are all roles within Windows Server 212, so there is no additional third-party software to buy. Windows Server 212 offers many important benefits: Efficient management Remote Desktop Services (RDS) in Windows Server 212 provides a collection of virtual desktop features that are tightly integrated and can be easily configured and managed by using the new centralized and unified management console and wizards. Customers can quickly and easily deploy pooled (single, shared master virtual machine) or personal (separate virtual machine instance for each user) virtual desktops by using the new scenario-based deployment tools. In addition, customers can automate the process of patching both pooled and personal virtual machines while minimizing the impact on resource use and bandwidth. Best value for virtual desktops RDS supports various lower-cost storage options, eliminating the need to rely on expensive SANs. Customers can reduce costs by conveniently storing virtual machine files (including configuration, virtual hard disk files, and snapshots) in shared folders. In addition, client virtual machine images can be hosted on server message block shares or Cluster Shared Volumes (CSVs) and cached on direct attached storage (DAS), enabling the best performance at lower cost. Rich experience everywhere Customers can enjoy the benefits of several user-experience enhancements to RemoteFX. These include 3-D graphics and streaming video across local area networks and wide area networks, expanded USB device redirection, a CPU-based graphics accelerator that allows applications running in a virtual machine to have an accelerated DirectX graphics experience, as well as full support for touch-enabled devices and applications. With the new User Profile Disk feature, it is possible to store user personalization and application cache data for pooled virtual machines and sessions, so data is maintained across user logon sessions. In addition, dynamic allocation of CPU, network, and disk I/O are balanced, preventing one user from impacting the performance of other users on the same server. For more information, see Microsoft Virtual Desktop Infrastructure. Fully tested, ready to run The design of the HP CV SMB RA for Windows Server 212 is based on extensive scalability testing conducted by a team from Microsoft in Redmond, Washington. The final design was then re-tested to ensure that the reference architecture meets all Microsoft and HP performance requirements. The sections that follow provide details of the tests. Performance testing software The initial scale tests and final reference architecture tests were performed using the Login VSI version 3.6 (V3.6) performance and scalability testing tool to generate desktop workloads and gather data. The tests used the standard medium user workload profile (equivalent to a knowledge worker), which consists of typical desktop activities such as using email; running office applications such as word processing, editing spreadsheets, and creating slideshows; and viewing high-definition (HD) video that is running on the host server. The test framework measured and reported the active (or logged-in) session response time for activities such as opening applications and user interface clicks, as well as CPU utilization, disk I/O latency, and queue length in pending I/O operations. For more information, see Login VSI. 4
Initial scalability tests The purpose of the initial scalability tests was to examine the limits of the HP and Microsoft VDI solution, and then to identify the sweet spot, or maximum number of user logins meeting HP and Microsoft performance requirements as defined by a CPU utilization of less than 8 percent and login response time of less than 3 seconds. This number of user logins was then used to design the final reference architecture. Initial scalability test environment For the initial scalability tests, the team tested two variations of target virtual machines: Windows 7 target virtual machines and Windows 8 target virtual machines. Both variations used Windows 7 client virtual machines and RDP 7. Table 1 shows a summary of the environment used in the initial scalability tests. Table 1. Initial scalability test environment summary Test rig and workload Launcher specifications User profiles Session Desktop virtual machines Login VSI V3.6 15 second interval parallel launching 8 virtual launchers (2 vcpus and 8 GB RAM) Single user profile was configured across all virtual machines to reduce the impact of IOPS and push the limit of the processor Resolution 124x768 No connection broker was used Sessions did not use network level authentication (NLA) Windows 7 with Service Pack 1 (SP1), 32-bit, 1 vcpu, dynamic memory (768-1536 MB, 2 percent buffer), 1 NIC with Windows 8 integration components installed Windows 8, 32-bit, 1 vcpu, dynamic memory (768-1536 MB, 2 percent buffer), 1 NIC Table 2 shows the configuration of the VDI host server used in the initial scalability testing. Table 2. VDI host server specifications HP ProLiant DL38p Gen8 Server 2 CPUs Intel Xeon E5-268 @ 2.7 GHz 256 GB RAM 16 GB DDR3 DIMMs 1333 MHz 8 drives 2 x 2 GB SSD SAS, RAID1 (used for operating system and golden image) 6 x 3 GB 15K RPM SAS, RAID5 (used for differential virtual hard disks [VHDX]) 4 x1 Gb Ethernet connections (one was used in tests) 5
Response time (in ms) CPU utilization (in percent) Initial scalability test results The team from Microsoft tested up to 25 simulated users on a single host. They then identified the number of user logins at the established limit of 8 percent CPU utilization and 3 second response time, and compared the results to find the sweet spot to use for the final reference architecture. The scalability test results were very impressive. The Login VSI limit for the tests running Windows 7 target virtual machines was reached at 239 users and at 232 users for the tests running Windows 8. The following sections detail the results of the tests. Results for tests with Windows 7 desktop virtual machine The following graphs show the results of the tests that used target virtual machines running Windows 7. Figure 1 shows the CPU utilization (or percent total run time) along with the number of user logins, over time. At the 8 percent CPU utilization limit, there were approximately 18 simulated users logged in to the system. Figure 1. CPU utilization for tests with Windows 7 target virtual machines 1 25 225 8 2 175 6 15 125 4 1 75 2 5 25 : :4 :8 :12 :17 :21 :25 :3 :34 :38 :43 :47 :51 :56 1: 1:4 1:9 1:13 1:17 1:22 1:26 1:3 Time (in HH:MM) Percent total run time Users Figure 2 shows the disk I/O latency with the number of user logins, over time. Figure 2. Disk latency for tests with Windows 7 target virtual machines 3 27 24 21 25 225 2 175 I/O read latency (ms) I/O write latency (ms) Users 18 15 12 9 6 3 15 125 1 75 5 25 : :4 :8 :12 :17 :21 :25 :3 :34 :38 :43 :47 :51 :56 1: 1:4 1:9 1:13 1:17 1:22 1:26 1:3 Time (in HH:MM) 6
Response time (in ms) Queue depth (pending I/Os) Figure 3 shows the queue length in pending I/O operations with the number of user logins, over time. Figure 3. Average disk queue length for tests with Windows 7 target virtual machines 2 18 16 25 225 2 Average disk queue length Users 14 12 1 8 6 4 2 175 15 125 1 75 5 25 : :4 :8 :12 :17 :21 :25 :3 :34 :38 :43 :47 :51 :56 1: 1:4 1:9 1:13 1:17 1:22 1:26 1:3 Time (in HH:MM) Figure 4 shows the response time with the number of user logins, over time. Note that the test measures and averages the first few user login sessions to create an initial baseline; in this test, the baseline was 876 milliseconds (ms). The maximum number of users reached was 239. At the 3 second limit, the number of simulated users reached was 23. Figure 4. User response time with the number of user logins, over time, for tests with Windows 7 target virtual machines 1 9 8 7 6 Minimum response Average response Maximum response VSI Index Average 5 4 3 2 1 25 5 75 1 125 15 175 2 225 25 7
Response time (in ms) CPU utilization (in percent) Results for Windows 8 desktop virtual machine The following graphs show the results of the tests that used target virtual machines running Windows 8. Figure 5 shows the CPU utilization along with the number of user logins over time. As with the Windows 7 tests, at the 8 percent CPU utilization limit, there were 175 simulated users logged in to the system. Figure 5. CPU utilization with Windows 8 target virtual machines 1 9 8 7 6 5 4 3 2 1 : :4 :8 :12 :17 :21 :25 :3 :34 :38 :43 :47 :51 :56 1: 1:4 1:9 1:13 1:17 1:22 Time (in HH:MM) 25 225 2 175 15 125 1 75 5 25 Percent total run time Users Figure 6 shows the disk I/O latency with the number of user logins, over time. Figure 6. Disk latency for tests with Windows 8 target virtual machines 1 9 8 7 25 225 2 175 I/O read latency (ms) I/O write latency (ms) Users 6 5 4 3 2 1 15 125 1 75 5 25 : :4 :8 :12 :17 :21 :25 :3 :34 :38 :43 :47 :51 :56 1: 1:4 1:9 1:13 1:17 1:22 Time (in HH:MM) 8
Response time (in ms) Queue depth (pending I/Os) Figure 7 shows the queue length in pending I/O operations with the number of user logins, over time. Figure 7. Average disk queue length for tests with Windows 8 target virtual machines 2 18 16 25 225 2 Average disk queue length Users 14 175 12 1 8 6 4 2 15 125 1 75 5 25 : :4 :8 :12 :17 :21 :25 :3 :34 :38 :43 :47 :51 :56 1: 1:4 1:9 1:13 1:17 1:22 Time (in HH:MM) Figure 8 shows the user response time with the number of user logins, over time. The baseline response time was 845 ms, and the maximum number of users reached was 232. At the 3 second limit, the number of users reached was 223. Figure 8. Login VSI user response time with the number of user logins, over time, for tests with Windows 8 target virtual machines 1 9 8 7 6 Minimum response Average response Maximum response VSI Index Average 5 4 3 2 1 25 5 75 1 125 15 175 2 225 25 Recommended reference architecture tests The results of the initial scalability tests were used to identify the maximum number of user logins meeting HP and Microsoft performance requirements. This sweet spot was the lower number of user logins that met both CPU utilization and response time requirements: 175 users. The recommended reference architecture uses two VDI host servers, a single connection broker, Windows 8 desktop virtual machines, and RDP7. A 35 virtual desktop pooled collection was created with 175 users per host server. This pool was set to utilize the new User Profile Disk feature in Windows Server 212, which maintains data across user logon sessions. NOTE: There are different methods of determining the recommended number of users for a server. Microsoft uses CPU utilization. HP looks at 6-65% of the VSImax value. This creates a recommended range from 155-175 users. 9
Reference architecture test environment Table 3 shows a summary of the environment used in the reference architecture tests. Table 3. Reference architecture test environment Test rig and workload Launcher specifications User profiles Session Desktop virtual machines Login VSI V3.6 5.14 seconds launch interval using parallel launching* 8 virtual launchers (2 vcpus and 8 GB RAM) Different users were used to connect to different virtual machines, and the user profile was not preloaded; profile disk was set to 5 GB (dynamic) Resolution 124x768 NLA was used for the sessions; connection broker was fully utilized to route users to the corresponding virtual machines Windows 8, 32-bit, 1 vcpu, dynamic memory (768-1536 MB, 2 percent buffer), 1 NIC * This login rate was derived from the 9:am logon storm, which assumes that all 35 employees will log on to their virtual machines within a 3 minute time period at the start of the work day. Table 4 shows the configuration of the VDI host servers used in reference architecture tests. Table 4. VDI host server configuration HP ProLiant DL38p Gen8 Server 2 CPUs Intel Xeon E5-268 @ 2.7 GHz 256 GB RAM 16 GB DDR3 DIMMs 1333 MHz 8 drives 2 x 2 GB SSD SAS, RAID1 (used for operating system and golden image) 6 x 3 GB 15K RPM SAS, RAID5 (used for differential virtual hard disks [VHDX]) 4 x1 Gb Ethernet connections (one was used in tests) In final reference architecture tests, the user data and profiles resided on a separate file share. Because using the local storage of the DL38p Gen8 server can affect the space available and the number of supported users, a file share should be used for storing user data and profiles to obtain optimal performance. Different users were used to connect to different virtual machines and the profile was not preloaded. The remote desktop connection broker used was located on a virtual machine (4-core, 8 GB RAM). Moreover, NLA was used for the sessions and the connection broker was fully utilized to route users to the corresponding virtual machines. 1
Response time (in ms) Reference architecture test results Figure 9 shows the user response time with the number of user logins, over time, for the recommended reference architecture. Note that the baseline for the reference architecture was 82 ms. Figure 9. Login VSI user response time with the number of user logins, over time, for the recommended reference architecture 4 375 35 325 3 275 25 225 2 175 15 125 1 75 5 25 25 5 75 1 125 15 175 2 225 25 275 3 325 35 Minimum response Average response Maximum response VSI Index Average Reference architecture bill of materials The reference architecture requires Windows Server 212 and the hardware listed in Table 5. Note that the table lists the hardware required to configure a single server. Note also that part numbers are at time of publication and subject to change. The bill of materials does not include complete support options or other rack and power requirements. Please consult with your HP product sales representative for more details. Table 5. Rack and power bill of materials Qty Part Number Description 1 6532-B21 HP DL38p Gen8 8-SFF CTO Server 1 662228-L21 HP DL38p Gen8 E5-268 FIO Kit 1 662228-B21 HP DL38p Gen8 E5-268 Kit 16 64791-B21 HP 16GB 2Rx4 PC3L-16R-9 Kit 2 65378-B21 HP 2GB 6G SAS SLC 2.5in SC EP SSD 6 652611-B21 HP 3GB 6G SAS 15K 2.5in SC ENT HDD 1 631681-B21 HP 2GB FBWC for P-Series Smart Array 1 68421-B21 HP Ethernet 1GbE 53FLR-SFP+ FIO Adptr 2 656363-B21 HP 75W CS Plat PL Ht Plg Pwr Supply Kit 1 663478-B21 HP 2U SFF BB Rail Gen8 Kit 2 53746-B21 HP NC112T PCI Express Gigabit Server Adapter Optional if redundant networking is required 11
Conclusion Together with Microsoft software and HP ProLiant DL38p Gen8 servers, the HP CV SMB RA for Windows Server 212 delivers an impressive VDI user density per server along with the many benefits of a traditional enterprise VDI without its complexity and startup cost a very practical VDI solution for small and medium-sized businesses. For more information To read more about HP and client virtualization, see hp.com/go/cv To read more about Windows Server 212, see microsoft.com/en-us/server-cloud/new.aspx To help us improve our documents, please provide feedback at hp.com/solutions/feedback. Call to action Contact your HP representative or HP reseller partner today to engage and discover how you can benefit from this client virtualization reference architecture. Achieve greater user mobility, anytime, anywhere access, and efficient user management all without the cost and complexity of traditional enterprise VDI. Get connected hp.com/go/getconnected Current HP driver, support, and security alerts delivered directly to your desktop Copyright 212 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice. The only warranties for HP products and services are set forth in the express warranty statements accompanying such products and services. Nothing herein should be construed as constituting an additional warranty. HP shall not be liable for technical or editorial errors or omissions contained herein. Microsoft and Windows are U.S. registered trademarks of Microsoft Corporation. Intel and Xeon are trademarks of Intel Corporation in the U.S. and other countries. 12 4AA4-391ENW, Created September 212