On 01/24/2012 07:28 AM, Dennis McCarthy wrote:
Hi Joel,
We do NOT have any PAV's. We are a pretty small shop. One production LPAR and
one test (sandbox for me). The VSAM file in question is open to a single CICS
region. Give that additional information, can I expect a negative impact on
response time going to the MOD-9's?
Dennis
You have the potential for a negative impact, but it all depends on the
related transaction rates and physical I/O rates in the CICS region. If
you have RMF or some other measurement tool that will give average
device busy on the current 3330-3's used by the file, you can get some
idea in the extreme cases: If average device busy is consistently below
5%, then I would think odds are pretty good that the negative impact
would be in acceptable ranges; if you frequently see some average device
busy of 20% or higher on multiple drives I would consider the odds high
for significant negative impact. In between, things are less clear. RMF
device measurement interval relative to arrival pattern of transactions
may be an issue -- if transactions are typically clustered in a small
part of the RMF measurement interval rather than uniformly distributed,
you may need a smaller interval to see if device usage could be a problem.
If you have CICS measurement tools of some kind, you may be able to see
other things of use, like number of logical I/Os against the file that
don't even result in physical I/O because of in-memory buffers in CICS,
and proportion of reads versus writes (which must eventually do a
physical write). With the major load coming from CICS, one of the
possibilities if you have real memory to spare is that you may be able
to compensate for any bottlenecks in physical I/O by throwing a large
number of additional LSR buffers at the file to raise the odds of
finding records in memory and reducing the number of logical reads that
require a physical read. Considering the size of the file, you
obviously can't have a significant amount of the file in buffers, so the
success of this strategy would depend on there being some pattern of
clustering in the way records are typically accessed.
So, if you're lucky, you may be in a zone where rule of thumb may
suggest a Yea or Nay; otherwise, there may be no simple way to determine
without taking more risk and trying it.
Assuming your device busy rates are in the grey zone, if real memory and
adding LSR buffers is an option and I had what I considered really valid
reasons why I really want to get to the 3390-9, I think I would try a
temporary experiment to significantly increase LSR buffers and see if
that significantly reduces I/O rate to the file. (You wouldn't want to
run in this mode too long on 3390-3's, because if it does help response,
the end users could get used to it and expect it). This way you could
at least prove in advance whether an LSR buffer increase could be used
to offset a negative impact if the migration to 3390-9 has more impact
than acceptable -- and that might impact your decision on whether to
attempt the migration.
That's the long answer. The short answer is as always "it depends".
--
Joel C. Ewing, Bentonville, AR jcew...@acm.org
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to lists...@bama.ua.edu with the message: INFO IBM-MAIN
