# of VMs per lun is something you should do the math on as the queue depth is a shared resource per lun.
Benchmarks on my system led me to believe that my LUNs get no performance increase beyond 29ish queued IOs.... so a I prefer a 32 io queue depth as opposed to the new 64 default in ESX 5.x. However, my SAN hardware is older and uses 4gb ports, so doing your own testing is always recommended.
Using 32 queued IOs/lun as our starting point. From there we figured 10-16 VMs per data store was a sweet spot... multiplied by our average VMDK size... we went with a default of 1 TB datastore. We do have some 2 TB datstores for special one-off needs. Also it is really easy to grow a datastore/vv if you start your vv off too small.... so that said, I think its best to start small and adjust bigger as needed.
_________________ Richard Siemers The views and opinions expressed are my own and do not necessarily reflect those of my employer.
|