David, would it be possible for you to collect some performance data on
the i/o improvements you're seeing there? Perhaps some CP MONITOR data
that we could analyze? It would be nice to know if PAV is really
something that Linux on z/VM sites should really be taking advantage of.
What was that saying again? "Can't measure it; not interested"?
Something like that from somebody on the West Coast....that ring any
bells? ;-)
DJ
David Kreuter wrote:
With PAV support for attached/dedicated devices you can achieve I/O
concurrency. I have been running some tests from linux guests for
simple I/O driving and the results with PAV are most impressive. I
have gotten as far as 1 base and 2 aliases so far (I'll keep
increasing the # of aliases until break even point). This is with
z/VM 520 and SUSE SLES9.
You need to do some setup in IOCP, z/VM, and linux (EVMS). Works
well.
I look forward to seeing this soon with my Oracle servers. David
-----Original Message----- From: The IBM z/VM Operating System on
behalf of Raymond Noal Sent: Tue 4/25/2006 1:45 PM To:
[email protected] Subject: [IBMVM] z/VM I/O Concurrency
Dear List,
I was following a topic thread on the IBM-MAIN list server where
someone wanted to create multiple Linux LPARs instead of running
Linux under z/VM. One respondent stated that z/VM only allows one I/O
to a "disk" at a time. Is this really true (Allan??). Does one I/O
per disk only apply to mini disks or to attached disks as well? I
find it hard to believe that z/VM would be this restrictive.
Your thoughts.
HITACHI DATA SYSTEMS
Raymond E. Noal Lab Manager, San Diego Facility Office: (858) 537 -
3268 Cell: (858) 248 - 1172
