IBM SmartCloud Virtual Desktop Infrastructure for Microsoft Windows Server 2012 VDI Reference architecture

Transcription

1 IBM SmartCloud Virtual Desktop Infrastructure for Microsoft Windows Server 2012 VDI 24 July 2013 Kent Swalin Copyright IBM Corporation, 2013

2 Table of contents Introduction...1 Architectural overview...1 Component model...2 Component relationship diagram... 2 Storage model... 4 Operational model...5 Deployment diagram... 6 High availability (HA)... 8 VDI Management servers / Virtualization host servers...8 Networking... 9 Storage integration Sizing considerations System configuration for 600 pooled users Appendix 1: Bill of materials...15 Appendix 2: Performance testing results...17 Performance results for virtualization host configuration Processor performance graph Average disk queue length graph Average disk transfer graph Memory- available bytes graph Login VSI VSImax graph for one of the virtualization host nodes Performance results for virtualization host configuration Processor performance graph Average disk queue length graph Average disk transfer graph Memory-available bytes graph Login VSI VSImax graph for one of the virtualization hosts Resources...26 Trademarks and special notices...27

3 Introduction This document describes the reference architecture for IBM SmartCloud Virtual Desktop Infrastructure with Microsoft Virtual Desktop Infrastructure (VDI) running on Microsoft Windows Server It should be read in conjunction with the IBM SmartCloud Virtual Desktop Infrastructure reference architecture document available at: ibm.com/partnerworld/page/stg_ast_eis_sdi_infrastructure The business problem, business value, and requirements are described in the IBM SmartCloud Desktop Infrastructure reference architecture and are not repeated here. This document gives an architecture overview and component model of the Microsoft pooled VDI and then describes the deployment model of Microsoft pooled VDI for varying numbers of users accessing virtual Windows desktops, running on servers that can be accessed from a variety of user devices such as tablets, thin clients, physical desktops, and notebooks. The intended audience of this document is IT professionals, technical architects, sales engineers, and consultants to assist in planning, designing and implementing Microsoft VDI on IBM System x 3550 M4 or IBM System x3650 M4 and IBM Storwize V3700. This instance of the architecture is based on a pooled virtual machine (VM) configuration. Pooled VMs are non-persistent user desktops. Every user VM has a dedicated portion of the virtualization host server s resources to guarantee the performance for each desktop. The desktop VM is dedicated to one user while the user is logged on. When the user logs off or restarts, the VM is returned to the pool and reset to the pristine gold image state for the next user. Microsoft has complementary technologies: Microsoft Remote Desktop Session Hosting (RD Session Host) and RemoteFX which can be combined with VDI. However, these technologies are not covered in this reference architecture. Architectural overview (IBM SmartCloud Virtual Desktop Infrastructure for Microsoft VDI) is an end-to-end desktop solution that lowers the barriers of adoption, such as cost, complexity, and coverage that are traditionally associated with the VDI technology. It allows access to a personal desktop from anywhere, using any device. Microsoft Windows Server 2012 is an ideal choice for desktop virtualization. It includes everything needed to set up a VDI solution including: the connection broker, web access, gateway, and other necessary components, so no additional third-party software is required. Figure 1 shows the main features of the IBM SmartCloud Virtual Desktop Infrastructure for Microsoft VDI reference architecture. 1

4 Figure 1: SmartCloud Virtual Desktop Infrastructure reference architecture for Microsoft VDI This document limits the discussion to the components inside the customer s intranet. This reference architecture does not address the issues of remote access and authorization, data traffic reduction, traffic monitoring, and the general issues of multisite deployment and network management. Component model Component relationship diagram Figure 2 is a layered view of the IBM SmartCloud Virtual Desktop Infrastructure mapped to the Microsoft VDI with Windows Server 2012 Hyper-V hypervisor. 2

5 Figure 2:IBM SmartCloud Virtual Desktop Infrastructure mapped to the Microsoft VDI The main components for Microsoft VDI using Hyper-V are explained in the following table: Virtualization Host Remote Desktop Virtualization Host integrates with Hyper-V to deploy pooled or personal virtual desktop collections. Remote Desktop Web Access Server Remote Desktop Gateway Server Remote Desktop License Server Remote Desktop Web Access enables users to access Desktop Connection through the Start menu on a computer that is running Windows 8, Windows 7, or through a web browser. Remote Desktop Gateway enables authorized users to connect to virtual desktops on an internal corporate network from any Internet-connected device. Remote Desktop (RD) Licensing manages the licenses required to connect to a virtual desktop. You can use RD Licensing to install, issue, and track the availability of licenses. 3

6 SQL Server SQL Server is used to store provisioning and broker connection information. SQL server is a required component. SQL Server 2012 standard edition or higher is recommended. Remote Desktop Connection Broker Server Clients Remote Desktop Protocol (RDP) Shared storage File Services Remote Desktop Connection Broker provides access to virtual desktops in a virtual desktop collection. It enables administrators to evenly distribute the load among RD Session Host servers in a session collection or pooled virtual desktops in a pooled virtual desktop collection. Users can access their virtual desktop from any device supported by the respective desktop virtualization solution; this includes company notebooks, home PCs, thin -client devices or tablets. IBM does not prescribe any particular approach for clients. Customers can repurpose existing desktops (which is typical for many deployments) or greenfield with thin- or zero-client devices. No agent is required on either the client or VM. The virtual desktop image is streamed to the user access device using RDP. Shared storage is required for this architecture. It is used to hold cluster information for the VDI management cluster, SQL data required for the VDI management servers, user data, and the VDI management virtual hard disks. The File Services role will allow the management cluster to present the user data as virtual hard disks to the pooled VMs Table 1: Main components of Microsoft VDI using Hyper-V Storage model This section describes the different types of data stored for this reference architecture. Pooled VMs use local storage for the client VMs. The pooled desktop enables users to connect to a new or different desktop image every time they log on while keeping aspects of user experience persistent. This allows the usage of local storage for the VMs instead of shared storage as no user-associated data resides persistently in the image. In case of a host failure, users can just reconnect to a desktop hosted on another system without the need for the VM to fail over to another node. The Windows Server OS and golden images are stored locally on the virtualization host on fast storage. The local storage on the virtualization hosts is summarized in the following table. 4

7 Volumes Use File format OS / Golden image Windows 2012 host OS, local golden image NTFS VMs Pooled VM Differential disks NTFS Table 2: Virtualization host storage Shared storage is used as a repository for the management of VMs, management databases, user profile and user data, and the client VMs master golden images. Whenever the golden master images are updated, they are then transferred to each virtualization host. The following table outlines the data layout for shared storage. Volumes Use File format Management Management VMs NTFS Management databases SQL server data NTFS User profile data User profiles and user data NTFS Master golden images Master gold images NTFS Table 3: Shared storage Operational model The pooled operational model is presented in this section. In order to illustrate the operational model for different sized customer environments, three different models are provided for supporting 150, 300, and 600 users. The VDI management cluster servers and virtualization host servers use Hyper-V as the hypervisor. The VDI management servers, which are clustered with Windows Server 2012 clustering, have management VMs instead of client VMs. The management VMs are shown in Figure 3. 5

8 Deployment diagram Remote Desktop Web Access 1 Remote Desktop Gateway 1 Remote Desktop Web Access 2 Remote Desktop Gateway 2 Management Cluster Remote Desktop Connection Broker 1 Remote Desktop License Server Remote Desktop Connection Broker 2 SQL 1 File Services SQL 2 File Services Virtualization Host Servers Remote Desktop Virtualization Host Remote Desktop Virtualization Host Remote Desktop Virtualization Host Remote Desktop Virtualization Host Figure 3: Server deployment diagram The VDI management servers have the File Services role installed to provide access to user profile data virtual hard disks to the virtualization host servers. This enables the virtualization host servers to share access to the user profile virtual hard disks as they reside on a cluster shared volume (CSV) disk on the shared storage. 6

9 The following table summarizes the VM requirements for each management VM. Management server VM Virtual processors Memory Storage Windows OS HA available Remote Desktop Web Access Server Remote Desktop Gateway Server Remote Desktop License Server SQL Server (Standard Edition or higher) Remote Desktop Connection Broker Server 2 4 GB 15 GB 2012 Yes 2 4 GB 15 GB 2012 No 2 4 GB 15 GB 2012 No 4 8 GB 15 GB 2012 Yes 2 8 GB 40 GB 2012 Yes Table 4: Management VM requirements The following table shows examples of typical client VM configurations. OS Version Number of processors Memory Network interface card (NIC) Mircrosoft Windows 7 (SP1) Microsoft Windows 8 32-bit or 64-bit MB Dynamic memory 32-bit or 64-bit MB Dynamic memory 1 1 7

10 Table 5: Pooled client VM configuration High availability (HA) High availability protects all layers of this solution. This was accomplished with: Two top-of-rack switches to provide redundancy Hyper-V clustering in the Management layer to provide high availability Multiple servers are used for high availability in the Virtualization host layer In this architecture, the virtualization hosts are not clustered as the Microsoft VDI collection is capable of spanning multiple virtualization host servers. In the rare event of a virtualization host failure, users on that host would need to reconnect to a new VM. The VDI management servers have a similar hardware specification as virtualization host servers do, so you can use them interchangeably in a worst-case scenario. If a management server goes down, it can be replaced by one of the virtualization host servers. VDI Management servers / Virtualization host servers At the core of the IBM SmartCloud Virtual Desktop Infrastructure for Microsoft VDI reference architecture, the IBM System x3550 M4 (1U) or IBM System x3650 M4 (2U) servers deliver the performance and reliability required for virtualizing business-critical applications in Hyper-V VDI environments. In this architecture, storage exists on local hard disks on the virtualization hosts. Two volumes are used per virtualization host. The first volume is configured using RAID 1 over two 200 GB solid-state drives (SSDs) and is used for the Windows Server 2012 host OS and the local golden image. The second volume is configured using RAID 0 and is used to host the provisioned desktop VMs along with their respective write caches. Pooled VMs use local storage and require the local storage on the virtualization hosts as shown in the following table. Volumes Size Number of disks RAID Use OS / Golden image 200 GB 2 1 Windows 2012 host OS, local golden image VMs 1-3 TB 6 or more 0 Pooled VM Differential disks Table 6: Virtualization host storage layout Testing has shown that the local VM volume needs at least six 15k drives to support 100 to 150 users with good performance. To support more users or if the I/O per user is higher, then eight drives have been 8

11 found to be effective, especially, if combined with higher powered processor and more memory. The I/O performance is shown in Appendix 2: Performance testing results. Networking Combinations of physical and virtual isolated networks at the host, switch, and storage layers meet isolation requirements. At the physical host layer, there are four 1Gb Ethernet devices for each Hyper-V server. The servers and storage maintain connectivity through multiple iscsi connections using multipath I/O (MPIO). Windows Server 2012 NIC teaming on the virtualization hosts provides fault tolerance to the host management and VM communication networks. At the physical switch layer, virtual local area networks (VLANs) provide logical isolation between various storage and data traffic. A key element is properly configuring the switches to maximize available bandwidth and reduce congestion. Based on individual environment preferences, you have choices regarding how many VLANs you create and what type of role-based traffic they handle. After you make a final selection, ensure that the switch configurations are saved or backed up. The configurations of the five VLANS are described in the following table. VLAN ID Name Description VLAN 10 iscsi storage network Used for iscsi storage traffic VLAN 20 iscsi storage network Used for iscsi storage traffic VLAN 30 Cluster private network Used for private cluster communication and cluster shared volume traffic VLAN 40 Cluster live migration network Used for cluster VM Live Migration traffic VLAN 50 Public network Used for host management and VM communication Table 7: Implemented VLANs At the physical storage layer, the IBM Storwize V3700 iscsi ports are used for connectivity. Each controller has two 1Gb Ethernet ports for iscsi traffic. The use of the IBM Subsystem Device Specific Module (SDDDSM) manages the multiple I/O paths between the host servers and storage, and optimizes the storage paths for maximum performance. VLANs are used to isolate storage traffic from other data traffic occurring on the switches. VLAN 10 and VLAN 20 are reserved for server access to the iscsi storage. All iscsi traffic should be isolated on VLAN 10 and VLAN 20. One switch hosts VLAN 10, and the second switch hosts VLAN 20. Implementation of the VLAN isolation from the storage controller: To help balance iscsi workloads, each Storwize V3700 controller maintains two iscsi connections to the networks. One connection from each controller to each switch. 9

12 Implementation of the VLAN isolation from each physical VDI management host: Each physical VDI management host will have two connections to the iscsi networks (one to each VLAN). Traffic isolation occurs at the switch. Storage integration The IBM Storwize V3700 system is used for shared storage which will host the management server VMs, the master golden image, SQL files, and user profile data. The IBM Storwize V3700 system uses iscsi ports for storage connectivity to the management cluster. Each controller has two 1GbE Ethernet ports for iscsi traffic. The IBM Storwize V3700 system was chosen for this implementation as it is an entry-level disk storage system designed with sophisticated capabilities unusual for a system of this class. It offers efficiency and flexibility through built-in thin provisioning and non-disruptive migration of data from existing storage. Built upon the innovative technology in the Storwize family, Storwize V3700 addresses block storage requirements at an affordable price. The following table outlines the volume requirements for shared storage. The number of disks is based on 300 GB 10K drives. Larger disk sizes could be chosen to meet the capacity needs of the client. The 24 disks in a single V3700 enclosure in the following table are based on a 600-user configuration. The number of disks can be reduced depending on the number of users expected. Refer to the Sizing considerations section for more information. Volumes Size Number of disks RAID Use Management 500 GB 6 5 Management VMs Management databases User profile data Mater golden images 200 GB 4 5 SQL Server data 2 TB 10 5 User profiles 500 GB 4 1 Master Gold Windows 8 or Windows 7 images Table 8: IBM Storwize V3700 storage layout 10

13 Sizing considerations One of the key questions when planning a VDI deployment is the number of users supported per host. Determining the system configuration able to support the load generated by users is a typical challenge. VDI requires a separate VM and associated hardware capacity for every user. Although one VM can host a relatively lightweight application (such as a data-entry application) that users access infrequently and with low resource costs, another might host a demanding computer-aided design (CAD) application requiring a lot of processor, RAM, disk, and network bandwidth. The test team developed two virtualization host configurations; the first is IBM System x3550 M4 or System x3650 M4 with 192GB of memory, two SSDs, six 300 GB 15k HDDs, and two Intel Xeon processors E5-2670; the second configuration is IBM System x3650 M4 with 256 GB of memory with two SSDs, eight 300 GB 15k HDDs, and two Intel Xeon processors E The two virtualization host configurations are designed to provide the flexibility for the client s requirements. The first configuration provides high performance with a reasonable price. The second configuration provides more processor power, memory, and I/O if the clients VMs have a more demanding workload. You can find the performance characteristics for these two configurations in the Appendix 2: Performance testing results section. These two configurations are shown in Table 9 and Table 10. By looking at these configurations, you can deploy a VDI solution that balances performance, price, failover, and scalability concerns. Virtualization server configuration 1 Server options Description Server Processors IBM System x3550 M4 or IBM System x3650 M4 Two Intel Xeon Processors E U or 2U Rack Server 20M Cache, 2.60 GHz, 8 cores Memory 192 GB Twelve 16GB 1.35v DDR3 RDIMMs OS disk / Golden image Two 200 GB SSD RAID 1 VM differential disk Six 15k 300GB RAID 0 NICs Two internal 1GbE ports Teamed in Windows Server OS Windows Server 2012 Datacenter Table 9: Virtualization server configuration 1 11

14 Virtualization server configuration 2 Server options Description Server Processors IBM System x3650 M4 Two Intel Xeon Processors E U Rack Server 20M Cache, 2.70 GHz, 8 cores Memory 256 GB Sixteen 16 GB 1.35v DDR3 RDIMMs OS disk / Golden image Two 200GB SSD RAID 1 VM differential disk Eight 15k 300GB RAID 0 NICs OS Table 10: Virtualization server configuration 2 Two internal 1GbE ports Windows Server 2012 Datacenter Teamed in Windows Server The test team designed these configurations to allow for sufficient resources if one of the management servers or virtualization hosts fail. Configuration users 300 users 600 users Management servers Virtualization host servers Servers required Configuration 2 Management servers Virtualization host servers Servers required Table 11: General sizing guideline 12

15 The following table shows the Storwize V3700 shared storage disk configuration sizing guidelines. Storage volumes 150 users 300 users to 600 users Management VMs 4 6 Management databases 2 4 User profile data 6 10 Master golden images 4 4 Total disk drives Table 12: Storwize V3700 disk configuration System configuration for 600 pooled users Figure 4 shows the deployment diagram for an IBM System x3550 M4-based solution that can support 600 pooled users. It is an option to use IBM System x3650 M4 servers instead of System x3550 M4 servers depending on your needs or hardware availability. 13

16 Management Virtualization Hosts 2 x IBM x3550 M4 2 x Intel E GB 4 x 300GB 10K 6 x IBM x3550 M4 2 x Intel E GB 2 x 200GB SSD 6 x 300GB 15K Storage IBM Storwize V x 300GB 10K Network Juniper EX2200 Top of Rack Network Switches - 1GbE Figure 4: Physical system configuration for 600 pooled users The following table summarizes the servers and user configuration for the virtualization hosts during normal activity and after a host failure. Virtualization host server Normal operating mode Failover mode Server users 120 users Server users 120 users Server users 120 users Server users 120 users Server users 120 users Server users 0 users Total users 600 users 600 users Table 13: Virtualization host server user configuration 14

17 Appendix 1: Bill of materials The following parts list was generated with the online IBM hardware configurator. All the parts might not be available in all geographies. Single System x3550 M4 virtualization host configuration 1 Quantity Description Part number 1 IBM System x3550 M4 7914AC1 1 x3550 M4 PCIe Riser Card 1 (1 x16 LP Slot) A1HJ 1 x3550 M4 4x 2.5" HDD Assembly Kit A1HG 1 Select Storage devices - no IBM-configured RAID required IBM System x Gen-III CMA A229 1 x3550 M4 PCIe Gen-III Riser Card 2(1 x8 FH/HL Slot) A1HK 1 IBM System x3550 M4 2.5" Base Without Power Supply A1H3 1 IBM System x 750W High Efficiency A1H5 Platinum AC Power Supply 1 Addl Intel Xeon Processor E C 2.6GHz 20MB 115W W/Fan A2B5 1 x3550 M4 plus 4x 2.5" HDD Assembly Kit A1HN 1 IBM System x Advanced Lightpath Kit A2U GB (1x16GB, 2Rx4, 1.35V) PC3L CL9 ECC DDR3 1333MHz A1QT LP RDIMM 1 Intel Xeon Processor E C 2.6GHz 20MB Cache 1600MHz A2B2 115W 1 ServeRAID M5110 SAS/SATA Controller for IBM System x A1WW 1 ServeRAID M5100 Series 512MB Flash/RAID 5 Upgrade for IBM A1J4 System x 1 2.8m, 10A/ V, C13 to IEC 320-C14 Rack Power Cable IBM 200GB SATA 2.5" MLC HS SSD A2FN 6 IBM 300GB 15K 6Gbps SAS 2.5" SFF HS HDD A283 1 IBM System x Gen-III Slides Kit A228 1 IBM System x Power Supply Filler for x3550 M4 and x1200 Enclosure A1HF 1 x3550 M4 System Level Code A1HB 1 ServeRAID M5100 Series 875mm Flash Power Module Cable A22C 1 IBM System x3550 M4 Planar A1H9 1 Windows Server 2012 (5731-W12) Per 2 Processor Server Datacenter 5731W12 15

18 Single System x3650 M4 virtualization host configuration 2 Quantity Description Part number 1 IBM System x3650 M4 7915AC1 1 System x3650 M4 PCIe Riser Card 1 (1 x8 FH/FL + 2 x8 FH/HL Slots) A1JT 2 IBM 200GB SATA 2.5 inch MLC HS SSD A2FN 1 System x3650 M4 8x 2.5" HS HDD Assembly Kit A1JX 1 IBM System x Gen-III CMA A229 1 x3650 M4 PCIe Gen-III Riser Card 2 (1 x8 FH/FL + 2 x8 FH/HL Slots) A1JU 2 IBM System x 750W High Efficiency Platinum AC Power Supply A1H5 1 Addl Intel Xeon Processor E C 2.7GHz 20MB 130W W/Fan A1L3 1 ServeRAID M5110e SAS/SATA Controller for IBM System x A2N2 1 IBM System x Advanced Lightpath Kit A2U GB (1x16GB, 1.5V)PC CL9 ECC DDR3 1333MHz LP A2R1 HyperCloud DIMM 1 Intel Xeon Processor E C 2.7GHz 20MB Cache 1600MHz A1KS 130W 2 4.3m, 10A/ V, C13 to IEC 320-C14 Rack Power Cable IBM System x Gen-III Slides Kit A228 1 IBM System x3650 M4 Planar A1KH 1 x3650 M4 Plus 8x 2.5" HS HDD Assembly Kit with Expander A1JY 8 IBM 300GB 15K 6Gbps SAS 2.5" SFF HS HDD A283 1 Windows Server 2012 (5731-W12) Per 2 Processor Server Datacenter 5731W12 16

19 V3700 iscsi shared storage Quantity Description Part number 1 IBM Storwize V3700 SFF Dual Control Enclosure C GB 10K 2.5 inch HDD ACLJ Top of rack switches Quantity Description Part number 2 Juniper 24 Port 1Gb EX2200 Ethernet Switch for IBM System x A13L Appendix 2: Performance testing results The performance analysis of the reference architecture was carried out using Login VSI. Login VSI is a software tool to test the performance and scalability of VDIs. The workload produced by Login VSI for the current performance analysis effort was representative of a typical set of activities performed by a knowledge worker (the medium workload). Resource utilization on the virtualization host servers were monitored using Microsoft performance monitor. All testing was performed with Login VSI version 4.0. It should be noted that the Login VSI VSImax as calculated in version 4 is lower than VSImax for previous Login VSI versions to closer represent a more realistic number of users that the system can support. The tests in this document using Login VSI used the following target client configuration: Windows 7 Microsoft Office 2010 Adobe Reader 9.3 Doro PDF version v1.82 Adobe Flash 11 ActiveX h Sun Java 7 Update 13 Microsoft.Net 3.5 Configuration options: Set the tuning parameter on the Broker Server. Set -RDVirtualDesktopConcurrency to 5 (current max) This option sets create/update to five VMs at a time (per host) rather than one, which is the default. Performance results for virtualization host configuration 1 The following tests show the performance for 160 users logging on during a 48-minute window with a total run time of 60 minutes. 17

20 Processor performance graph The processor graph in the following figure shows the logical processor utilization during performance analysis of a single virtualization host server. The results show sustained logical processor percentage runtime peaking at approximately 90%. The average logical processor percentage runtime of about 80% is ideal for an environment that is operating at the optimal combination of maximizing density, while providing sufficient headroom to ensure that user experience is not diminished. Figure 5: Processor utilization during a 60-minute run with 160 users logging in over the first 48 minutes The next two graphs show two different disk performance measurements. The test team measured the average disk queue length and the average disk transfer. These were measured to verify that there were no issues on any of the local disk arrays. Average disk queue length graph The average disk queue length graph tracks the number of requests that are queued and waiting for a disk during the sample interval and the requests in service. In Figure 6, the disk queue length in red color in is the SSD RAID 1 array, and the line in blue color is the average disk queue for the local storage six-disk RAID 0 array that are used to hold the differential disks for the user VMs. The average disk queue length shows no bottlenecks on the disk drives. 18

21 Figure 6: Average disk queue length graph for 60 minutes Average disk transfer graph The following figure shows that the average disk transfer is well below the 20 ms which would indicate a bottleneck on the disk. Figure 7: Average disk transfer graph for a 60-minute run 19

22 Memory- available bytes graph The memory graph in the following figure shows the available memory during performance analysis. It can be seen that the available memory drops to approximately 10 GB, which shows that the system is getting close to being constrained on memory with 160 users logged on to the system, the total memory on the virtualization host is 192 GB. These VMs were configured to use dynamic memory; from a minimum of 512 MB to a maximum of 2 GB. After all the users have been logged on, the available memory returns to approximately 30 GB, which demonstrates the efficiencies provided by Microsoft dynamic memory technology. Figure 8: Memory-available bytes graph for 60 minute test run Login VSI VSImax graph for one of the virtualization host nodes The Login VSI tool generates a test score called VSImax after each successful test run. The VSImax is the maximum capacity of the tested system expressed in the number of Login VSI sessions. The VSImax score can then be used to determine the performance of a particular system configuration. This graph shows that according to Login VSI, this configuration supports 157 users per virtualization host with reasonable performance. The performance counter shows that 125 users on the System x3550 M4 server is probably more realistic as the processor utilization with 157 users is near 95%. 20

23 Figure 9: Login VSI VSImax graph for one virtualization host node Performance results for virtualization host configuration 2 The Storwize V3700 storage and network remain the same for this configuration. The management and virtualization host servers have been changed to use the IBM System x3650 M4 server in order to use more local disk drives for the virtualization host servers. In this configuration, there are two SSDs in a RAID 1 configuration, eight local disks in a RAID 0 configuration, 256 GB of memory, and Intel Xeon processors E The tests explained in this section show the performance for 205 users logging in during a 48-minute window with a total run time of 60 minutes. Processor performance graph The graph in the following figure shows the logical processor utilization during performance analysis of a single virtualization host server. The results show sustained logical processor percentage runtime peaking at approximately 90%. The average logical processor percentage runtime of about 80% is ideal for an environment that is operating at the optimal combination of maximizing density while providing sufficient headroom to ensure that user experience is not diminished. 21

24 Figure 10: Processor utilization during a 55-minute test run with 200 users logging in over the first 48 minutes. Average disk queue length graph The average disk queue length graph tracks the number of requests that are queued and waiting for a disk during the sample interval, and requests in service. In Figure 11, the disk queue length in red color is the SSD RAID 1 array, and the blue color line is the average disk queue for the local storage eight-disk RAID 0 array that are used to hold the differential disks for the user VMs. The average disk queue length shows no bottlenecks on the disk drives. 22

25 Figure 11: Average disk transfer graph for a 60-minute run Average disk transfer graph The graph in the following figure shows that the average disk transfer is well below 20 ms and this would indicate a bottleneck on the disk. Figure 12: Average disk transfer graph for a 60-minute run 23

26 Memory-available bytes graph The memory graph in the following figure shows the available memory during performance analysis. It can be seen that the available memory never drops below 80 GB, which shows that the system is not constrained on memory. The total memory on the virtualization host is 256 GB. The total memory for the VMs configured on this system is 200 GB and this demonstrates the efficiencies provided by Microsoft s dynamic memory technology. Figure 13: Memory-available bytes graph for the 60-minute test run Login VSI VSImax graph for one of the virtualization hosts This graph in the following figure shows that this configuration can support 184 users per virtualization host with reasonable performance. The performance counters show that 138 users on each System x3650 M4 is very realistic and that if a node were to fail the System x3650 M4 server can support 184 users. The tests show that these systems are well balanced in terms of processor, memory, and disk configuration. 24

27 Figure 14: Login VSI VSImax graph for one virtualization host node 25

28 Resources IBM SmartCloud Virtual Desktop Infrastructure reference architecture ibm.com/partnerworld/wps/servlet/contenthandler/stg_ast_eis_sdi_infrastructure Microsoft Virtual Desktop Infrastructure 26

29 Trademarks and special notices Copyright IBM Corporation References in this document to IBM products or services do not imply that IBM intends to make them available in every country. IBM, the IBM logo, and ibm.com are trademarks or registered trademarks of International Business Machines Corporation in the United States, other countries, or both. If these and other IBM trademarked terms are marked on their first occurrence in this information with a trademark symbol ( or ), these symbols indicate U.S. registered or common law trademarks owned by IBM at the time this information was published. Such trademarks may also be registered or common law trademarks in other countries. A current list of IBM trademarks is available on the Web at "Copyright and trademark information" at Microsoft, Windows, Windows NT, and the Windows logo are trademarks of Microsoft Corporation in the United States, other countries, or both. Intel, Intel Inside (logos), Xeon, MMX, and Pentium are trademarks of Intel Corporation in the United States, other countries, or both. Other company, product, or service names may be trademarks or service marks of others. Information is provided "AS IS" without warranty of any kind. All customer examples described are presented as illustrations of how those customers have used IBM products and the results they may have achieved. Actual environmental costs and performance characteristics may vary by customer. Information concerning non-ibm products was obtained from a supplier of these products, published announcement material, or other publicly available sources and does not constitute an endorsement of such products by IBM. Sources for non-ibm list prices and performance numbers are taken from publicly available information, including vendor announcements and vendor worldwide homepages. IBM has not tested these products and cannot confirm the accuracy of performance, capability, or any other claims related to non-ibm products. Questions on the capability of non-ibm products should be addressed to the supplier of those products. All statements regarding IBM future direction and intent are subject to change or withdrawal without notice, and represent goals and objectives only. Contact your local IBM office or IBM authorized reseller for the full text of the specific Statement of Direction. Some information addresses anticipated future capabilities. Such information is not intended as a definitive statement of a commitment to specific levels of performance, function or delivery schedules with respect to any future products. Such commitments are only made in IBM product announcements. The information is presented here to communicate IBM's current investment and development activities as a good faith effort to help with our customers' future planning. Performance is based on measurements and projections using standard IBM benchmarks in a controlled environment. The actual throughput or performance that any user will experience will vary depending upon 27

30 considerations such as the amount of multiprogramming in the user's job stream, the I/O configuration, the storage configuration, and the workload processed. Therefore, no assurance can be given that an individual user will achieve throughput or performance improvements equivalent to the ratios stated here. Photographs shown are of engineering prototypes. Changes may be incorporated in production models. Any references in this information to non-ibm websites are provided for convenience only and do not in any manner serve as an endorsement of those websites. The materials at those websites are not part of the materials for this IBM product and use of those websites is at your own risk. 28