Storage made simple, use the VDI Storage Decision Tree

I have lost track of how many articles about VDI and Storage I have written. It seems that no matter how many times I write about it people are still  confused about what type of storage technology to choose from when architecting VDI solutions. Yet this week I had to explain the difference between XtremeIO, Atlantis and FusionIO.

So, I decided to change the strategy and created what I like to call the “VDI Storage Decision Tree”. The concept is simple – follow the workflow tree and go talk to the recommended vendor or vendors. In some cases, there will be a single vendor to talk to, but other cases multiple vendors provide solution to the problem.

If you are a vendor and feel you are in the wrong bucket, or you feel that you should be in more buckets, or if you are not in any bucket, post a comment and we will fix it. Just tell me about your technology and why you should be in a given bucket.


I have moved this post to a dedicated page. Please find most up-to-date workflow at


Click on the picture below to enlarge.


This article was first published by Andre Leibovici (@andreleibovici) at


Skip to comment form

  1. Hey Andre,
    Great post. You may want to remove Tintri from the ‘All Flash Array’ section, they are an Hybrid Array. Nimbus Data on the other hand are all flash.

  2. Daryl, Thanks. I have updated the workflow.

  3. There are some small flaws in the diagram. In some paths questions eliminate themselves. The problem is the “do you want scale out” question after the initial “do you want to deploy VDI > NO” line 🙂

    To be complete you should also add SimpliVity and ScaleComputing in the HyperConverged group.

    Overall VERY GOOD diagram!?

    • Duncan on 08/16/2013 at 3:05 am

    I would have rather seen it a few extra levels deep. You know ask the “consumer” if they want a “hyperconverged” / “all-flash” / “hybrid” array, but this shouldn’t be a question by itself as the user shouldn’t really care what type of array it is, as long as it solves their problem.

  4. Duncan, I see what you are saying. However, at some point customers have to decide based on technology. As an example, a hybrid may be able to offer the same IOPs and throughput and a all-flash, depending of the configuration and the sizing required. Also, the buckets would get too big with multiple vendors; even thou their technology is completely different.

  5. Hans, thanks for your comment. I am adding SimpliVity and ScaleComputing in the HyperConverged bucket.

    On the other issue, it’s a way back only. I’ll add a little arrow negating using the line.

  6. Hi Andre,
    It’s interesting diagram to put together.
    Why is there two “do you want a scale boxes”? The second one (lower down) is redundant because the user has already said no. That means there shouldn’t be a choice other than cost or performance. So I’d consolidate those two into one decision – what is the primary factor: cost with the arrow going down or performance going to the all-flash question.
    It looks like you are try to push as many users towards a hyper-converged solution.
    I actually think the scale-out question is unnecessary and misleading. First, no one is going to say they want a solution that doesn’t scale. Second, I’m sure all the storage vendors would (somewhat) rightly argue that their solutions scale-out. I presume what you mean is that the storage doesn’t scale with servers in a linear, hyper-converged fashion. I guess what I don’t understand is why you differentiate between XtremeIO and other all-flash arrays in that regard?
    I think it’s valid to ask a customer if they want hyper-converged or not, as most people have opinions on that. But I’d suggest ditching the scale-out question altogether.

  7. The Scale Out question is very well in it’s place here as it is the question whether or not you want to scale in independent nodes instead of controllers + disk shelves.

  8. Forbes, thanks for your comment. I’m not trying to push anybody to any specific storage solution. I’m just trying to steer customers into the right direction for their needs, requirements and desires.

    XtremeIO is a scale-out solution, allowing you to add units and them being part of the same namespace. As far as I know this is not true for other vendors. Please, correct me is I am wrong.

    So, I take scale out as if it’s possible to add blocks and still be a single solution.

  9. While not hyper-converged, NetApp does offer a scale-out storage solution with Clustered Data ONTAP.

    Very good article. Really nice to see all of the many options presented this way. With this many options, the customer always wins!

  10. Andre – I’d frame the logic flow differently. In every VDI deal we deploy (disclaimer – I work for EMC/XtremIO) it comes down to five things.

    1. Driving an attractive overall $/desktop (not just storage, but all-in)

    2. Delivering the best end-user experience (better than a physical desktop)

    3. Making life easy for administrators and enabling them to be agile

    4. Complete freedom to choose full or linked clone VDI models with no drawbacks either way (rather than being forced into linked clone technology because of storage shortcomings)

    5. Knowing that the solution will work, not just in a pilot but in a scale deployment that may be thousands or even tens of thousands of desktops.

  11. Josh, you points don’t make sense. How to you categorize a solution as ‘deliver best end-user experience’. This is sales BS and all vendors will say the same. I am letting consumers know how to choose a solution/vendor based on technical requirements, then they go and talk to you if you fit their technical requirements.

  12. Forbes, you were actually right on the duplicate entry for “Scale-out”. I have updated the workflow. Thanks!

  13. Interesting. Where’s the I want to use local storage without adding unnecessary cost option? 🙂

  14. Michel, you should follow the path that leads to “Do you want to use existing shared storage?”, then follow “No”. This path will only be available for non-persistent desktops.

Comments have been disabled.