What's all this talk about VDI and IOPS?

It often comes as a surprise that VDI demands on storage performance can be far greater than with server virtualization.  When you think about it, it's pretty easy to understand why... When you login to your laptop, the hard disk drive light is glowing. It's swamped delivering I/O for just one user.  So when we load 50 or 100 users per server, and more per storage array, it's no surprise that the I/O demands exceed the ability of the array to keep up.  So how is it that a SAN array with 16 drives can keep up with hundreds of users when a single disk is swamped for a single user?

It's true that SAN storage is faster than desktop storage - and that helps a lot.  The typical SAN disk can deliver about twice the performance of a desktop disk drive.  Plus the array offers excellent caching - which helps some with read, but helps a lot with write (especially since RAID 5 or 6 can have a substantial penalty for striping).  So if you take 16 drive array (which typically becomes 14 drives after RAID), and you assume that each drive can handle twice the I/O of the desktop disks ... and we even give an extra 50% improvement for the caching - we can get about 3 users per drive - or 14 * 3 = 42 users.  But of course we can load up the array with more than 42 users, right?  That array can cost $50,000 or more!

And the answer is yes - but not as much as you might think.  The reason we can load more users on the array is because not all users are experiencing an I/O storm at the same time.  Therefore we don't need to design the storage array to handle a worst cast peak load that will never happen.  What it needs to do is handle a typical peak load - and that is the key measurement you need to understand.

There are tools that can help measure your existing IOPS load - which can be very helpful to build a profile of your storage needs.  So if you have the time and money, it is a worthwhile exercise.  Just be careful that you are measuring load that will be similar to the load you will have in VDI.  For example, VDI desktops are typically already booted in advance of use whereas desktops/laptops often boot in the morning.  And if you plan on using Application Virtualization / Streaming and/or a Profile Manager, those solutions can also add substantial I/O load on your storage (note that Unidesk does not need Application Virtualization or a Profile Manager, and as such there is no increased load, but that is another story...).

We generally find that an array of 16 drives (SAS 15K RPM with RAID 5/6) can deliver something around 100-200 VDI users.  SSD (solid state, or memory based disks) have screaming fast IOPS - but at a substantial price tag per GB.  Some vendors offer a middle-ground solution - a hybrid array that blends SSD and SAS using a load balancer that moves data between the two efficiently. These hybrid arrays can be a great alternative when combined with Unidesk, as we substantially reduce the capacity footprint (commonly 70-80%) and deliver an I/O pattern that is easily load balanced by the array.  Another great opportunity to reduce the cost of storage per user is to consider local storage in addition to, or potentially instead of, SAN.  The cost difference is substantial, but as with everything in life, there are tradeoffs. 

I'll be writing a bunch more about how to configure, optimize and reduce the cost of storage, including using technologies like hybrid arrays and local storage, in upcoming blogs.  In the meantime, feel free to contact me or the Unidesk team for assistance in designing a well performing, cost effective VDI solution.