Dynamic Disk Provisioning…

Have been tracking a few internal discussions concerning dynamic disk provisioning – the idea of being able to add disk capacity to the storage supporting your Exchange infrastructure at any time.  Now the way I understand it; and I imagine vendors and hardware platforms will do this in many different ways, is that larger LUNs are presented to the OS than is actually available in physical disk capacity.  If you ever want to make use of the additional capacity physical disks are provisioned at that time.  In my experience it’s also one of the reasons that is put forward to differentiate a SAN solution from one based on DAS.  The majority of SAN vendors provide management tools that make online expansion very simple.

My preference for designing Exchange Servers is to design the server to support a specific maximum number of mailboxes; a figure that’s obviously dependent on a number of factors including user and usage profiles and maximum mailbox size.  This determines both the capacity and performance requirements of the storage supporting the server.  Once you’ve designed an appropriate solution, and tested it, then you know that the storage you have provisioned for your mailbox role servers will provide sufficient performance and capacity which in turn provide a suitable user experience.  Once in production you fill it up with mailboxes and when you need more provision a new server and storage of exactly the same specification. (…making sure you have sufficient CAS’s, HUB’s, GC’s etc.. to support the additional mailboxes of course.)

Once you start changing the physical hardware that you tested then it’s impossible to be sure that the solution will provide that suitable user experience.  And if you are adding capacity that means that something's changed. ..it means that the number of mailboxes is increasing, or that the amount of data per mailbox is increasing, or that perhaps that usage patterns have changed.  In all cases this changes the IOPS profile you are now required to support, which means you can make no guarantees about the solution and crucially the user experience as a result of the change to the storage.

I have also heard the argument that dynamic disk provisioning enables you to purchase only the storage that you require for the first deployment of mailboxes for example, or where you are expecting your user base to grow significantly.  Not sure this this is that practical.  Most server infrastructure has a lifecycle of typically about 3-4 years.  So in most cases it should be possible to design and provision a solution that lasts for that lifecycle.  Maybe that’s not realistic for every deployment but I think it should be the aim…