Building a worldwide scalable VMware View environment.

1. Building a worldwide scalable VMware View environment. By: Tim Brown 10/13/2009

2. Agenda: • VMware View Overview • Steps to get from a concept to a reality • Testing environment I used to Validate VMware View • Good Practices for building a VMware View environment.

3. VMware View Overview

4. Why VDI? • Business Continuity • Contractors • Keep IP safe • Cost savings • Power savings • Flexibility

5. VMware View Landscape

6. Steps to get from a concept to a reality

7. Phases of a Successful Deployment Project

8. Guess the Outcome? Phases Often Followed in Deployment Projects

9. Define • Who are you users? • What are your users doing? • Where are your users? • When are your users logging in? • Peak user connections • Specific Hardware needs? • How much printing do your users do?

10. Gather Information • Measure physical desktop usage: Understanding how different applications affect physical processor, memory, network and storage resources is of key importance when sizing the servers and storage. PERFMON can be used to collect required information on a Windows desktop. • Estimate processor requirements: To estimate processor requirements, Administrators can use PERFMON to measure average processor utilization for the physical desktop, then multiply the measurement by the targeted number of virtual desktops for each VMware ESX host. Other factors to consider include storage and network virtualization overhead, RDP protocol overhead, and incorporating sufficient headroom to account for period for spikes and redundancy. • Estimate memory requirements: Because memory usage can vary based on VMware ESX page sharing, calculating memory requirements requires some estimates. The high-water mark can be estimated by ignoring the effect of page sharing. For example, to create 64 virtual machines with 512 MB of RAM each, a hypervisor without page sharing would require at least 32 GB of RAM. With page sharing on the ESX host, actual memory requirements in production will be much lower. • Calculate network requirements: Dividing the estimated traffic by the speed of the network interface card (NIC) yields the number of NICs typically required. To avoid large queuing delays, it is better to size for about 70% utilization on the NICs (0.7 * NIC speed). Administrators should also consider the type of NICs used, multimedia applications, shared folders, and the connection broker protocol. • Estimate storage capacity and performance requirements: Consider both the I/O per second (IOPS) and storage throughput data gathered from the physical workload. Given the significant capacity of disk drives today and modest throughput requirements of many common VMware View workloads, in practice, administrators should give careful attention to creating a storage configuration with enough disk drives to meet the aggregate IOPS requirements.

11. Verify your Estimates • To simulate a real-world VMware View environment, You can use software from AutoIT (for more information, go to http://www.autoitscript.com) to simulate desktop workers running the applications and functions listed below and add any specific to your company discovered in the definition phase. You can write AutoIT scripts to randomize workload mimicking end user work patterns. • You configuration might involve the following typical business applications: • Microsoft Word (open, modify random pages, save and close) • Microsoft Excel (writing to cells, sorting, formulas, charts) • Microsoft PowerPoint (slideshow and edit slides) • Adobe Acrobat (Open and browse random pages) • Microsoft Internet Explorer (browse plain-text and web album) • WinZip (install/uninstall) • McAfee VirusScan (continuous on-access scan) • Every VMware View environment will have its own unique performance characteristics based on the specific end-user requirements. There is no “one-size-fits-all” configuration for servers and storage.

12. Design a VDI Scalable Server/Storage Block • Using these numbers you need a server/storage platform that delivers these measured values. • Once you exhaust your Server/Storage block simply replicate what you’ve done for a simple to manage solution.

13. View Composer Shining Knight or Trojan Horse? VMware View Composer is one of the new components included in VMware View 3 and it provides the following benefits when deployed: • Reduces storage costs from 40 to 90%. • Reduces desktop provisioning from 15 minutes to just seconds. • Manages hundreds of desktops from a single central image, retaining user settings when updating or patching service packs, application updates or even OS upgrades. • Rolls back instantaneously, enabling customers to streamline management and guarantee that all user systems are up to date. • Reduces the number of images to manage. Please Note: It should be recognized that these technologies need to be applied carefully, as they potentially incur an impact to performance, by inducing a greater load on a more limited set of shared storage resources. A classic storage sizing tradeoff exists between configuring storage to optimize cost and capacity at the expense of performance.

14. Other Things to Consider… • Does my Storage vendor offer de-duplication/thin provisioning? Perhaps you choose to use those technologies instead of using View Composer. • Do you want all your users to be on the same patch level? Is there need to have dissimilar patches on desktops? • What if one user has a problem with a certain patch and needs it rolled out? • Guest OS patching: Patching all the virtual desktops on a single LUN at the same time can cause a spike in disk I/O to that LUN. Consider staggering patch rollouts or performing them during off hours or maintenance windows. • Manual virus scans: The ability for users to initiate their own virus scans can cause increased disk I/O, especially if multiple users are doing these scans simultaneously. • Desktop search tools: Desktop search tools installed inside the guest OS can also cause increases in disk I/O. • De-fragmentation: If users have the ability to run their own disk defrag during normal business hours, increases in disk I/O can result, affecting all users sharing the same LUN for their virtual desktops. • Cold boot, suspend/resume operations: Consider the effects if all users start work at the same time and all boot their desktops simultaneously. Suspending desktops is preferred over cold shutdown as the desktop can resume from suspend operations much quicker with less impact on the disk subsystem.

15. “To thin client or not to thin client? That is the Question” • Usually thin clients are solid state. (no moving parts) • About the cost of a PC but much longer life span. • Can be managed/upgraded remotely. • If stolen there is no IP on the unit. • Usually uses an embedded/hardened OS to cut down on exploits.

16. Network Performance Tips • Implement WAN Acceleration technology (If applicable) • During configuration disable SSL from VMware View for maximum performance. (Offload SSL to your Load Balancer) • Disable RDP Encryption/Compression if using WAN Acceleration • Think about deploying QOS on your network

17. Testing environment I used to Validate VMware View

18. Non-Accelerated Environment

19. Non-Accelerated Testing Results • I started < 10ms latency the result was it was just like a desktop everything behaved as designed including USB redirection/720p Video(WMV)/MP3 audio (flash is quite poor even at that speed) • Once we got in to ~30 ms latency the desktop still fully usable 720p video/mp3 audio worked but then USB redirection became slower. • Once you got into ~50ms latency the 720p video began dropping frames/MP3 audio was good the USB redirection worked but became very slow. Web surfing/Typical office work was still good • At 100ms 720p video did not function MP3 was fine USB was very slow desktop still useable • At 175ms MP3 audio was fine USB worked but really beyond practical. Desktop = Usable • At 225ms same results as 175ms • At 275ms MP3 audio still was good the desktop was a bit laggy I probably wouldn’t want to use it 8 hours a day but it really wasn’t bad…

20. Accelerated Environment

21. Accelerated Testing Results • < 10ms latency same as unaccelerated • At 30ms everything but USB redirection looked local • At 50ms everything but USB redirection looked local including 720p video and audio • At 100ms same results as 50 ms • At 175ms same results as 50 ms • At 225ms same results as 50 ms • At 275ms same results as 50 ms • I decided at this point I better check the WAN simulator I ended up dropping down to T1 this HD video was still playing without dropped frames. • I finally dropped it to a 128k ISDN link and then it finally broke so my WAN simulator was working as expected… • I also decided to go for broke and dialed in 400ms latency and had the same results as 50 ms

22. Good Practices for building a VMware View environment.

23. General View Good Practices • Slim down View Desktop guest OSs • Install and enable only the services and components needed • Nlite and Vlite serve as good examples of possibilities www.nliteos.com, www.vlite.net • Avoid over-provisioning memory and vCPUs • Size Windows page file between 50-100% of memory • Use folder redirection for My Documents Size the Parent VM System Disk appropriately • Replica creation time greatly affected by Parent VM disks

24. General View Good Practices • Validate size of User Data Disk • Provisioning too little disk for UDD can be detrimental • Outlook .OST will likely end up on UDD in Outlook cached mode (and UDDs are difficult to resize) • Validate the cost and performance tradeoffs of Cached Outlook mode for VDI • Adjust browser cache size to lowest useful setting • Do not suspend VMs on idle (Timeouts can occur when too many VMs simultaneously resume)

25. Questions?