ADSM-L

Re: Server Abending

1998-08-04 10:00:19
Subject: Re: Server Abending
From: Bill Colwell <bcolwell AT DRAPER DOT COM>
Date: Tue, 4 Aug 1998 10:00:19 -0400
In <85256656.00449391.00 AT csc DOT com>, on 08/04/98
   at 08:34 AM, Virginia Hysock <Virginia_L_Hysock AT CSC DOT COM> said:

>OS/390 2.4 MVS Server, just converted to 3.1.1.3, with a mirrored 18 GB
>database AIX, HPUX, Windows, NT, Netware clients (300+)

>Hi all,

>     I finally upgraded my ADSM server over the past weekend from 2.1.13 to
>3.1.1.3.  Everything was looking great until my first set of scheduled
>backups kicked in Monday night.  My server abended several times with
>various messages, including ANR0529W Insufficient Memory on the server.  I
>have my buffpoolsize set to 24576, my logpoolsize set to 8192 and
>maxsessions set to 325.  The messages manual recommends decreasing all of
>these settings and/or increasing the memory allotted to the server.  My MVS
>tech says that 128M (which is recommended in the Quick Start Guide) is a
>lot already.

>     Anyone else ever experience a similar problem after upgrading?  I am
>hoping to resolve this before too many days go by, in case I am forced to
>back out the new level.  TIA,

>                              Ginny

I wonder what is taking up all your memory?  what is the difference between
v2 & v3 for memory allocation?

The ANRSERV load module is 2 megs larger.

Logpoolsize of 8192 sounds excessive.  I run with 1024 and the 'Log Pool
Pct. Wait' value shown by 'q log f=d' is always 0.00.  The books say to
monitor this to set the option correctly.

It is easy to understand the effect of changing the poolsize numbers but
there is no doc that says the cost of each session in maxsessions. Hey IBM,
how can we tune this value to solve memory problems when we don't know the
memory cost of a session?  This is what appendix A says to do and we are
flying blind!

325 sounds excessive when you have 300 clients. Why is it so high? I have
1100 clients with maxsessions set to 200.

I would try changing maxsessions and logpoolsize first, and then the jcl
region size.  I have found a large buffpoolsize is critical for performance.


--
-----------------------------------------------------------
-----------------------------------------------------------
Bill Colwell
Bill Colwell
C. S. Draper Lab
Cambridge, Ma.
bcolwell AT draper DOT com
-----------------------------------------------------------
=========================================================================
<Prev in Thread] Current Thread [Next in Thread>