is VSAN more affordable?

I was very pleased to see actual results coming on the VMware blog concerning VSAN 5.5 GA benchmarks in the lab. First of all please read the blogpost first before going to my analysis/questions. Secondly if you are looking for answers, this is not going to be the place as I am just going to ask more questions.

That being said let’s have a look at the actual results. If we simplify the results of the benchmark these are the finding:

  • VSAN does scale lineair, up to 1600 heavy duty VDI’s on a 16-node cluster
  • response times scale linear as well (equal no matter what the size of the cluster)
  • The tested, average DELL R720 server backed VSAN with today’s CPU’s scales at a 100:1 consolidation ratio

Virtualsan55 scaling

Affordable or not?

Now I have been out of real life pre-sales for a while so what I have here are speculation and guessing. But the benchmark did however give us an idea of the used hardware so I can make a guess at what this would cost. I have tried to simplify only the EXTRA costs to run a VSAN comparing to an external array. The price of the server, networking and vsphere licenses (incl vCenter) are not to be discussed here as they would be the same for an external array.

Virtualsan55 setup

  • 32 VSAN licenses at $2.500 = $ 80.000
  • 32 x 200GB PCIe at $1.250 = $ 40.000
  • 192 x 15k HDD at $ 250 = $ 48.000
  • TOTAL cost: $ 168.000

You’ll see I actually made the 15k disk as cheap as possible because there are so many that it could inflate the cost tremendously. I encourage the VMware benchmark team to give us a complete bill of materials so we could make a true costs comparison. The first question that pops to mind is: is a $100/VDI comparable to a Nimble Storage / Tintri / PureStorage / … solution?

Apples and Oranges

Do you remember my Fooled by Marketing Metrics series? Remember what I said about unit/dollar metric? When buying array/appliance based hardware the affordability only counts on it’s full potential. So … if one of those boxes I previously named actually would only cost $150.000 which would be about 10% cheaper, this would only stay cheaper than a VSAN if you use it to it’s full force. Why? Because VSAN is a modular system and if I would only need 800 VDI’s you could scale it down to 8 nodes and buy 1 extra per 100 seats. That doesn’t count for the arrays. They would become double as expensive for smaller environments.

Advantage to modular systems: buy what you need, when you need it!

To truly compare apples to apples we should add the cost of the server hardware + vSphere licenses and compare to a Nutanix/Simplivity or even a Maxta/HP Lefthand VSA/… solution. They are the only ones that can truly be compared to a modular system like VSAN from an economics perspective.

It’s all worthless

Now to make this all go bad: the whole benchmark is most probably going to be worthless in the end. I can hardly imagine that the saturation of this 1600 VDI solution is storage. It’s most definitely going to be CPU/MEM bound. The direct result of this is that we are definitely WASTING RESOURCES. I don’t think we need 32 PCIe SSD’s + 192 15k disks to power a 1600 VDI environment. I guess if you put 32 PCIe SSD’s in an external array you’d get a lot more VDI’s out of it. This begs the question on the affordability of a VSAN solution again.

Questions

  • VMware: can we have a bill of materials?
  • VMware: have you monitored saturation/usage of the CPU/MEM/Flash/Disks ?
  • Array vendors: what’s the VDI mark number of your solution against what price?
  • Hyper-Converged vendors:  what’s the VDI mark number of your solution against what price per node?

I want to do this is all openness so I encourage anyone involved to reach out to me and make this comparison.

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.

7 comments

  1. Good thoughts Hans, I had something similar in mind for this and a few other solutions.

    I think it comes down to picking the right option for your requirements and design for it. The flexibility of scaling on something like VSAN may be just what someone needs.

    On the other hand modern flash based arrays may prove to offer better performance and certainly features right not. And for customers that start with a large number of desktops and scale in bigger chunks this may prove to be the right choice.

    It’s good to see VMware offer this option and competition for hardware vendors is good. Force them to innovate and drive down costs.

  2. You don’t need PCIE SSDs or 15K drives; it just happened to be the config of the hardware that was used for this benchmark. You can pretty much get the same result with Intel S3700 SSDs and decent SAS drives.

    1. Hi Captain,

      I agree and you confirm my point that for this case VDI is wasting resources as the saturation is happening on CPU. You’ll run the same amount of VDI’s on a flash enhanced array with the same virtualization hosts. Now if you look a the prices I choose, you’ll notice that you might win a few $K with the smallest intel S3700 cards available though not that much for enterprise 10k SAS disks as I did make them cheap as hell not to inflate the message.

    1. Well, as mentioned before I am no longer actively involved in pre-sales so I had to choose cheap averages. In real life current PCIe Devices would be more expensive as they tend to be bigger than 200GB as well. I made very careful assumptions to not inflate the price so that the conversation here would not lead to the fact that I would exaggerate on purpose.

Leave a Reply to Brian Cancel reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.