RE: Why not *SHRPOOL1?

["Ingvaldson, Scott" <SIngvaldson@xxxxxxxxxxxxx>]

The basic reason for multiple pools is to reduce the competition between
different types of workloads, especially long running batch workloads, short
high CPU interactive transactions, and the system jobs running in the *Base
pool. 

QPFRADJ takes time to adjust because it only moves a small amout of memory
at a time to avoid overcompensating.( I'm pretty sure that it's every 30
seconds, someone correct me if I'm wrong)  On the other hand, it always
overcompensates after moving from a day type workload to a night type
workload, i.e. 95% interactive to 95% batch; it will move as much memory as
it can to the active pool and this can cause some applications to experience
long startup times as QPFRADJ takes it's time moving the memory back where
it is needed.

You can mitigate this somewhat by setting minimum/maximum sizes using
WRKSHRPOOL and F11 to adjust the tuning data.  In your case, if you set the
minimum size of pools 3 and 5 to 20% they would not drop below about
3000(M), though they would still be able to swap the additional memory
between the pools as needed.

You can find more info about this in Software Knowledgebase document
#4621424, Separating Batch Work from *BASE at
http://www-912.ibm.com/s_dir/slkbase.nsf/slkbase

Regards,
Scott Ingvaldson
AS/400 System Administrator
GuideOne Insurance Group

-----Original Message-----

Date: Mon, 27 Jan 2003 09:10:24 -0500
From: rob@dekko.com
Subject: Why not *SHRPOOL1?

On our 840 12-way we have system value QPFRADJ set to "2=Adjustment at IPL 
and automatic adjustment".  And we have the following pools:
System    Pool    Reserved    Max 
 Pool   Size (M)  Size (M)  Active  Pool 
   1     2618.98    655.60   +++++  *MACHINE 
   2     6307.05     55.12     530  *BASE 
   3     3935.35      <.01      80  *SHRPOOL1
   4      303.44       .00      20  *SPOOL 
   5     1835.15       .20     227  *INTERACT

Actually, this does a pretty good job.  However, if you start this big 
whopping job in one subsystem, or the other, then it takes awhile until 
QPFRADJ switches memory over.  In the Rochester lab (just last week) we 
had a Query take 3 minutes until we switched QINTER to run in *SHRPOOL1. 
They were still scratching their heads to figure out how often QPFRADJ 
adjusts.  After the switch it took only a very few seconds.

So now my boss asked me a good question.  Why don't we just use two pools: 
 *MACHINE and *SHRPOOL1?  And avoid that time lag of QPFRADJ.

Questions and comments appreciated.

Rob Berendt