Translating Max IOPS into how many users you can fit onto a storage array(s) can be a pretty complicated question to ask of someone. Workloads are random between users, different blocks sizes are hitting the storage, AV is getting in the way and the list goes on and on. 20 IOPS per user is big joke in the VDI space. While 20 IOPS may represent the average, the deviation from that number can be astonishing. I encourage you watch Applied Math for VDI Design: A Statistical Approach to Designing VDI Environments. The session was presented at BriFourm last year and now is free. If you watch the video, you will see and understand what you need to be planning for.
VDI-IOmark from Evaluator Group is a tool that can tests your storage subsystem with realistic workloads. IOmeter can only give you simulated IO and not all the crazy behaviors of actual workload. VDI-IOmark uses workload replay from previous work captured from VMware RAWC implementation(View Planner). VDI-IOmark has 64 unique replays. The workloads range from 5 IOPS -20 IOPS on average but have peaks over 100. Microsoft Office(Excel, Powerpoint and Outlook), Internet Explorer, 7zip and Windows Media Player were all used to creat the replay. Boot and steady were also included in the replay. The tests are not dependent on your server platform, switching fabric or storage protocol so that’s a big thumbs up but it doesn need to run on .
The great thing about VDI-IO mark it that requires less the time to configure than building out a full environment. This is great if want to repurpose some old storage and test it our prior too or if you’re a consultant, you can do apple to apple comparisons between different vendors. Also since is each replay file contains 8 workloads test you can test your storage subsystem will less server hardware, you don’t need all the RAM it would normally take in a traditional LoginVSI or View Planner test. It’s always hard to get more money for test gear but people always want to know what the expensive box can do in the datacenter.
VDI-IO mark is still not a free pass for full VDI solution. VDI-IO mark doesn’t test the capacity of your servers(CPU, memory), network and the efficiency of your hypervisor. You will still need to use View Planner or LoginVSI to validate everything. You also don’t have the chance to add in your own applications which could be a big deal in some environments. It would also be great in future product releases if could they take a stab at test splitting the IO based on the View Composer architecture.
What I really do like is having the ability run tests on a storage subsystem today and then months later test some new gear and see how they compare knowing that the tests are repeatable. It will be interesting to see how Evaluator Groups keeps old replays around for such comparisons as the product evolves.
Some other interesting things about VDI-IO Mark
To test a 256 user workload you need:
79 GB of RAM
9 cores at 3 Ghz
4 Gb\s to your storage subsystem.
Currently VDI-IOmark is only for VMware View.
Cost for end users (not vendors) is $2,500 per year for unlimited use
and includes support. Free for 30 days. *Price is also negotiable*
Vendors, resellers, VAR’s etc pricing is additional.
The below graphs show the randomness of one user workload that was captured. It’s very interesting to see the different IO amounts as compared to the amount of data being transferred. It really shows why you need a specialised tool for benchmarking the number of VDI users that your system can handle.
For another take on VDI-IOmark you can read: