RE: Performance of large LF access path builds - SSD vs 10K spinny

rob@xxxxxxxxx wrote on Tue, 16 Sep 2014 11:57:02 GMT:

Also related, suppose you're an all SSD shop. And all your
disks are hosted on IBM i. And that IBM i has multiple lpars
it hosts (IBM i and AIX). "IF" IBM i was altered in 7.2 to
automatically sense that you are running on SSD's would that
auto sense pick up, even on a guested lpar, on the fact that
even though the disks are hosted by another lpar that the
hosting lpar is all SSD's and it should increase the thread
support?

I believe that 7.1 TR5 in the hosting LPAR is when we started
passing information to client LPARs the NWSSTG was or was not
backed by SSD. An earlier TR (or maybe it was base 6.1, I
forget), would recognize of VSCSI units were SSD backed or not.

If not, if I specify UNIT SSD on the guests would it
use the increased thread support? Would using the increased
thread support be a bad thing because it's now guested?

If 100% of the hosting LPAR is SSD or if 100% of the NWSSTGs are
mapped across SSDs, then yes with 7.1 changing to designate the
table or index is on SSD may utilize an increased number of
tasks and improve performance.