A KAMINARIO WHITE PAPER VMWare Horizon View Solution Guide August 2013
Table of Contents Executive Summary 3 Why Kaminario K2 4 Test Environment 4 Test Pool Configuration 5 Test Image Configuration 6 Management Test Results 7 Desktop Deployment Times 7 Power-on/Suspend/Resume 1,000 VMs 7 View Planner Test Results 8 Stage I: Dedicated K2 8 Stage II: All-Purpose K2 10 Detailed Test Results 11 Conclusion 12 VMWARE HORIZON VIEW SOLUTION GUIDE AUG 2013 2
Executive Summary VMware Horizon View is one of the two leaders in the VDI (Virtual Desktop Infrastructure) space a space created by VMware itself over half a decade ago with its product of the same name (VMware VDI). Implementing VDI has the potential of huge ROI IT departments gain the ability to manage desktops centrally, with tremendous economies of scale through drastically reduced management overhead: Hardware-specific Golden Images are a thing of the past the number of images to manage and their complexity is reduced significantly since there s only one platform to create for: VMware Virtual Hardware Patching and Upgrading OSs and Applications becomes much easier and faster Increased control means securing the desktops becomes also much less of a burden, especially if vshield-enabled security solutions are deployed BYOD (Bring Your Own Device) policies are no longer a burden to IT departments there are VMware Horizon View clients for virtually any hardware platform, including any platform that can run an HTML5 browser Unfortunately, this very compelling story has a snag: although the ROI is great, a VDI implementation can be complex, and, if the wrong hardware is used, become in itself expensive and a nightmare to manage and support. In particular, many VDI implementations have been abandoned because of either end-user uproar due to poor desktop performance, or IT departments giving up due to too much complexity. Fortunately, Kaminario s K2 All-Flash Array can solve all these problems, and help create affordable, easy to implement, scalable, easy-to-manage-and-support VDI deployments that will have end-users and IT departments alike raving about the huge improvement in productivity. In this document, we will show how this is possible, and provide you with real-world test results to back up our lofty claims. VMWARE HORIZON VIEW SOLUTION GUIDE AUG 2013 3
Why Kaminario K2 With CPU power and RAM space being as abundant (and underutilized) as they are today, in most virtualization projects IT departments find that the bottleneck is their legacy storage. In short: Virtualization is Storage-bound. Why? Well, by nature virtualizing servers and desktops shuffles I/O even more, generating highly random, unpredictable workloads that choke both 100% HDD-based and hybrid storage. Sequential speeds are meaningless. Caches are useless. With Kaminario K2, you can finally virtualize your desktops without fear of end-user riots. The K2 is so fast with random, mixed workloads, the end-users will actually get a better storage experience in a virtual desktop running on a K2 than they would with a physical desktop or laptop. The K2 handles massive Boot storms without skipping a beat. Boot times will be spectacular, even if 10,000 users are all booting at the same time. And on top of all of that you don t need to dedicate a K2 for VDI; you can run your Virtual DB Servers and your Virtual desktops on the same K2: we can handle it, no problem. If that s not enough, K2 also provides other dramatic management advantages: Provisioning and Rollout are extremely fast; as we ll see in the test results, VMs can be cloned at speeds that were thought to be impossible until the K2 came along Daily VM operations boot, suspend, resume are done at lightening speeds No need for additional 3rd party tools and appliances because K2 is all flash and Scale out, there is no need to worry about multiple management consoles, caching appliances, or any other add-ons that increase complexity and cost Optimization of VMware Horizon View s own management tools K2 is best deployed with Linked Clones, which leverage all of View s desktop management technologies, and is VAAI-enabled. Test Environment One Kaminario K2 2 K-Block flash array with 24TB of usable physical capacity The Cluster is comprised of 16 Dell M610 blade servers installed in single Chassis plus one 2U Dell R810 server. All the servers were running ESXi version 5.1. The ESX hosts were connected to the K2 via Brocade 8GB FC switches. Each FC port on each host was mapped (using zoning) to 4 ports on the K2. The Dell R810 server was running infrastructure VMs: One View Connection Server Single vsphere vcenter Server View Composer AD View Planner test appliance Win7 32-bit View Planner Test Image Golden Desktop VMWARE HORIZON VIEW SOLUTION GUIDE AUG 2013 4
TEST POOL CONFIGURATION For the Linked Clone testing, the following settings were used: POOL PARAMETER POOL TYPE USER ASSIGNMENT DESKTOP POWER POLICY AUTOMATIC LOGOFF DISPLAY PROTOCOL Setting Automated Pool Floating Take no action Never PCoIP MONITORS 2 MONITOR RESOLUTION 1920 x 1200 ADOBE FLASH QUALITY ADOBE FLASH THROTTLING DISPOSABLE DISK Do Not Control Disabled None NUMBER OF DESKTOPS 1,000 NUMBER OF DESKTOPS POWERED ON 1,000 PROVISION ALL DESKTOPS UP FRONT? LINKED CLONE OVERCOMMIT POLICY Yes Conservative DEDICATED REPLICA STORAGE None (replica in each clone data store) VMWARE HORIZON VIEW SOLUTION GUIDE AUG 2013 5
TEST IMAGE CONFIGURATION ATTRIBUTE DESKTOP OS SPECIFICATION Windows 7 Enterprise HARDWARE VMware Virtual Hardware version 7 VCPU 1 VMEMORY OS PAGEFILE 1024MB 1534MB starting & maximum VNICS 1 VIRTUAL NETWORK ADAPTER 1 VIRTUAL SCSI CONTROLLER 0 VIRTUAL DISK VMDK VIRTUAL FLOPPY DRIVE 1 VIRTUAL CD/DVD DRIVE 1 VMXNet3 Adapter LSI Logic Parallel 16GB Removed Removed VMWARE VIEW AGENT VMware View Agent 5.2 INSTALLED APPLICATIONS Internet Explorer 9 LINKED CLONE OVERCOMMIT POLICY Conservative DEDICATED REPLICA STORAGE None (replica in each clone data store) VMWARE HORIZON VIEW SOLUTION GUIDE AUG 2013 6
Management Test Results DESKTOP DEPLOYMENT TIMES Provisioning 1,000 Linked Clone desktops took 2 hours and 10 minutes. We felt those times were excessive, but after reviewing the statistics from the K2, we realized the load on the array was minimal. That points to the bottleneck being server processing power. We chose however not to add a more powerful server just for the purpose of speeding up initial desktop times, because that would be unrealistic: it s unlikely an IT department will size the vcenter server for initial deployment, since that would not generate good ROI. POWER-ON/SUSPEND/RESUME 1,000 VMS ACTION APPROX. TIME TO COMPLETE PEAK I/O BANDWIDTH MAXIMUM LATENCY Power-on 1000 VMs 3 minutes, 32 seconds 284,506 IOPS / 4.97GB/s Read 0.6ms Suspend 1000 VMs 4 minutes, 5 seconds ~35,000 IOPS / 9GB/s Write 2.98ms Resume 1000 VMs 2 minutes, 35 seconds 319,342 IOPS / 12.37GB/s Read 0.91ms VMWARE HORIZON VIEW SOLUTION GUIDE AUG 2013 7
View Planner Test Results STAGE I: DEDICATED K2 We wanted to start by seeing what kind of performance we would get if we dedicated a K2 to VMware Horizon View, with a Linked Clone deployment. To achieve a passing View Planner score (which translates into a good user experience), the latency values must be 1.5 seconds or less. The K2 enabled much better than that, as can be seen in fact, we set a new record at 0.51 seconds: Test Name: VPBig-Full-1000 Test Mode: local QoS Summary ----------- Group A : PASSED The 95th percentile was: 0.513599 seconds (To pass, this must not be more than 1.5 seconds) Workload Summary ----------------- Users: 1000 Iterations (Total): 5 (to pass, this must be at least 5) Iterations (Scored): 3 Workload Status: PASSED VMWARE HORIZON VIEW SOLUTION GUIDE AUG 2013 8
AdobeReader-Browse AdobeReader-Close AdobeReader-Maximize AdobeReader-Minimize Excel_Sort-Close Excel_Sort-Compute Excel_Sort-Entry Excel_Sort-Maximize Excel_Sort-Minimize Firefox-Close IE_ApacheDoc-Browse IE_ApacheDoc-Close IE_WebAlbum-Browse IE_WebAlbum-Close Outlook-Close Outlook-Maximize Outlook-Open Outlook-Read Outlook-Restore PPTx-AppendSlides PPTx-Close PPTx-Maximize PPTx-Minimize PPTx-ModifySlides PPTx-RunSlideShow Video-Close Word-Close Word-Maximize Word-Minimize Word-Modify Here are the detailed results: MEAN APPLICATION RESPONSE TIMES WITH 1,000 LINKED CLONES 0.00 0.17 0.34 0.51 0.68 0.85 1.02 1.19 1.36 1.53 1.70 Passing Mark VMWARE HORIZON VIEW SOLUTION GUIDE AUG 2013 9
LOAD ON THE KAMINARIO K2 DURING APPLICATION TESTING LOAD ON THE KAMINARIO K2 DURING BOOT STORM OF 1,000 DESKTOPS VMWARE HORIZON VIEW SOLUTION GUIDE AUG 2013 10
STAGE II: ALL-PURPOSE K2 As nice as the above results are, we noticed one thing: the K2 seemed to be almost idle. While running the View Planner test we are averaging under 5K IOPS and under 100MB/s that s a tiny fraction of what the K2 can do. We did break a record, but it was clear that it was mostly due to our extremely low latency we can handle a lot more IOPS and Bandwidth. So we decided to try something no other storage vendor has ever done: re-run the View Planner tests on a very, very busy K2 busy running both an OLAP and OLTP Oracle load. Essentially we were trying to simulate a situation where a customer is using a K2 as all-purpose storage running VDI, OLTP and OLAP loads in the same system, all at the same time. So we started: An OLAP load transferring 300M rows from a staging table to a fact table at changing degrees of parallelism (8-16 threads) in an Oracle session. An OLTP load created by 10-20 users randomly picking 0.02% primary keys from the above table and querying on them. These loads were run repeatedly during the entire duration of the following tests. and we re-ran View Planner while doing all of the above. The results? Test Name: VPBig-Full-1000 Test Mode: local QoS Summary ----------- Group A : PASSED The 95th percentile was: 0.513661 seconds (To pass, this must not be more than 1.5 seconds) Workload Summary Users: 1000 Iterations (Total): 5 (to pass, this must be at least 5) Iterations (Scored): 3 Workload Status: PASSED As you can see, the impact of running a massive OLAP/OLTP load is 0.000062 seconds 0.01%, which is actually within the margin of error for View Planner meaning K2 is truly an all-purpose array than can handle all your mission critical loads in one unit. VMWARE HORIZON VIEW SOLUTION GUIDE AUG 2013 11
AdobeReader-Browse AdobeReader-Close AdobeReader-Maximize AdobeReader-Minimize Excel_Sort-Close Excel_Sort-Compute Excel_Sort-Entry Excel_Sort-Maximize Excel_Sort-Minimize Firefox-Close IE_ApacheDoc-Browse IE_ApacheDoc-Close IE_WebAlbum-Browse IE_WebAlbum-Close Outlook-Close Outlook-Maximize Outlook-Open Outlook-Read Outlook-Restore PPTx-AppendSlides PPTx-Close PPTx-Maximize PPTx-Minimize PPTx-ModifySlides PPTx-RunSlideShow Video-Close Word-Close Word-Maximize Word-Minimize Word-Modify DETAILED TEST RESULTS MEAN APPLICATION RESPONSE TIMES WITH 1,000 LINKED CLONES 0.00 0.17 0.34 0.51 0.68 0.85 1.02 1.19 1.36 1.53 1.70 Passing Mark VMWARE HORIZON VIEW SOLUTION GUIDE AUG 2013 12
LOAD ON THE KAMINARIO K2 DURING VIEW PLANNER AND ORACLE LOADS VMWARE HORIZON VIEW SOLUTION GUIDE AUG 2013 13
Conclusion Our testing proved that the Kaminario K2 is truly the only all-purpose, all-encompassing All-Flash Array, the only option that will give companies one storage system for all their needs, with the fastest possible performance. The only no-compromises storage solution. VMWARE HORIZON VIEW SOLUTION GUIDE AUG 2013 14