Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss

TSO Performance in OS/390 R2

10 views
Skip to first unread message

Yair Elharrar

unread,
Dec 1, 1997, 3:00:00 AM12/1/97
to

Our old 4.3 system had a steady TSO response time of 0.07sec in the first
period during work hours. Ever since we moved to OS/390 R2, TSO response
time is jumpy - one day we have a perfect 0.07, the next day it's 2.67sec,
the next it's 0.35sec, then 0.08sec etc.

I calculate the daily average based on SMF record 72. I checked the records
and indeed they hold high values.

This happens both in Goal mode and in Compatibility mode (with the old IPS/ICS).

Any ideas what could be wrong??


TIA

Yair

Steve Samson

unread,
Dec 1, 1997, 3:00:00 AM12/1/97
to Yair Elharrar

Yair, does your ICS classify any system address spaces to control
performance groups? If so, you may have fallen into one of the
"improvements" of Version 5 and OS/390--they presume to BELIEVE you mean
that CONSOLE, for instance, should operate at a dispatching priority
below what it normally gets.

Scrub the ICS before your next IPL. RPGNs are OK, but not PGNs for
system spaces. Once the "system" attribute is lost, it's gone for the
life of the IPL. (It's fixed in OS/390 2.4.)

I hope this easy fix is the problem.

Steve Samson

Mark Zelden

unread,
Dec 2, 1997, 3:00:00 AM12/2/97
to Yair Elharrar

No way to know for sure, but I would guess the inconsistencies have
something to do with the I/O environment. Are all the correct libraries
in LLA/VLF (clists). At my current client , which has 200 TSO users on
one system and 100 on another - all going after the same ISPF libraries,
the biggest problem was device busys. I put ISPLLIBs and ISPPLIBs
in LLA/VLF, put clists libraries in the IKJEXEC VLF class, removed TSO
STEPLIBs (by putting them in the LNKLIST) and reduced first period
response times to under .10 seconds consistently. Period 2 averages
under .75 seconds. Before the changes first period had 2 or 3 second
response times spikes during prime time heavy usage. Another
contention problem was the SYS1.BRODCAST data set, which was (and
still is) on the same volume as some of the ISPF libraries, but
after the LLA/VLF changes there was no need to address this.

Hope this helps,
Mark

+--------------------------------+--------------------------------+
| Mark Zelden | Consultant |
| http://www.flash.net/~mzelden/ | CIBER Network Services, Inc. |
| mailto:mze...@flash.net | mailto:mze...@chi.ciber.com |
+--------------------------------+--------------------------------+
Check out my MVS utilities page at:
http://www.flash.net/~mzelden/mvsutil.html

0 new messages