IBM recommends using round robin for the XIV per best practice, which we have set already since all paths are active/active. I am reading over more documentation now and saw an interesting piece. By default, at least in ESX 4.x, round robin will use another path after every 1000 I/Ops, but lowering this to their recommended 10(!) should have a drastic improvement for latency, MBps and I/Ops.
They give a handy little script that will loop through all XIV paths and make this change. Of course I would need to run it on every hosts, and then again after adding another volume. I am thinking this can be better accomplished via the vMA to run the script against al hosts simultaniously or something...
But what are your thoughts on this change? No one from IBM ever mentioned this to me so this is quite new. We have a Gen 2 and two Gen 3 XIV's now with a couple more on the way...