We're about to transition to a new HP EVA4400 with Polyserve for our virtual machines. We picked this for 2 main reasons: The NAS heads are N-way active with all files available across all nodes, and the EVA storage controller stripes all data across every disk in the array to maximize performance (ideal for our 650 virtuals that all have unpredictable performance requirements). Since we have an entirely random load of 600+ virtuals and growing, we felt the best possible performance would come by giving evey IO possible to each machine rather than creating silos and hot spots in our data.
My question is, given that no matter how many volumes I make, all of the data is going to be striped across every disk and shared across every NAS head, are there any limitations inherent to NFS that would prevent me from just making giant volumes? Currently we've got a number of 1TB volumes, but between snapshots and thin provisioning I'm constantly having to tweak the size to keep up with the data growth.
If my disk performance and NAS performance is identical on twenty 1TB volumes vs one 20TB volume, is there any reason not to just do one big volume over NFS? Can I run 600-1000 virtual machines from one volume? Are there file locking issues or simultaneous connection issues or open file limitations using VSphere 4.x over NFS?
