Thank you for everyone that got in touch after my “Is VSAN affordable” post of a few days ago. I have enough new thoughts to give the post the extent it deserves and share a few new thoughts/questions.
First of all it has become pretty clear that my assumptions that the VSAN VDImark benchmark was CPU constraint appeared to be correct. The only true conclusion we can make from the original VMware post is that VSAN in G.A. has more CPU reduction than in beta which made them able to run about 5% more VDI seats on the same CPU’s.
The Economics of VSAN
But I am happy that it kicked of a lot of thoughts for me into the economics of VSAN as such. What I don’t know is what the VDImark result of that same host would have been on external storage. This would give us a total insight on the CPU-overhead of a VDI solution on VSAN.
Here’s a thought: if the CPU overhead for VSAN in comparison to external flash enhanced storage is big enough to run the same amount of VMs on one host less, is the total cost of HOST + VSAN licenses + VSAN hardware still worth it? In other words: is VSAN an economically acceptable solution to run a CPU constraint application.
It’s all in the design
So we have figured out by now that the 32 PCIe cards and 192 15k-disks was a huge overkill for this VDI solution. An Intel s3700 SSD and a few 10k or maybe even 7.5k disks would probably have been sufficient. This would already drive the cost of the necessary hardware down but still leaves room for thoughts towards the licensing per host in a CPU constraint environment.
The answer is: be smart with your design: if CPU is a constraint: scale your CPU’s! I don’t need to draw you a picture to tell you that you’ll run more seats on a 16-core processor comparing to an 8-core. And all this for the same VSAN license + VSAN hardware cost.
Food for thoughts
I am a huge fan of per node scalability. I really really don’t like scaling CAPEX upfront for future use like we used to do in 2 controller array environments with huge uplifts on upgrades and forklifts on migration.
However … in a world where we separate compute from storage, I can still scale independently on both sides. If I need more compute: add a server, if I need more storage (both performance and capacity) add flash/hdd disks. All will depend on the granularity of upgrade. Do I want to take the bullet of adding VSAN licenses and hardware when I only need more compute? What if I need a little more capacity, do I have to scale-up all nodes involved?
Of course this question is the same for all hyper-converged vendors. Don’t hesitate on chiming in folks 😉
Disclaimer: I am starting a consultancy project for ScaleComputing that is considered a VSAN competitor soon. This post is not guided by it other than the fact that this matter is top of mind for me right now.