Re: Why separate pools? -- MIDRANGE-L

A couple of notes that I think were not covered..

1) *CALC works well when you have similar jobs, using the same sets of
database files, and you have lots of memory in the pool. You have to keep
memory in the pool using the WRKSHRPOOL minimums, because the autotuner
basically cannot see the effect of the extra memory in use by *CALC and will
remove it without regard to *CALC.
When you are not using the same set of database files, *CALC will have
little effect - it is all about keeping records in memory (caching). When
you have a small amount of memory, *CALC will appear to have little effect.

2) Using shared pools is a great idea - be careful not to overdo it.
Separate your work into like jobs, and set them up to run in shared pools
together. Use maximum and minimum limits for advantage. Let the autotuner do
its work and watch where the limits are - for example, if it always tunes it
up to the maximum, increase the maximum %... if it always tunes to the
minimum - reduce the minumum % and so on... Pools like *SPOOL can often be
set to a minumum of 0.01% and will still work ok.

3) The autotuner sucks at tuning activity levels. The idea is to have the
activity level at the most efficient number - where the system is handling
the most number of jobs in the most effective way. You will find that basic
autotuning will just keep adding more and more activity levels until it has
"too many". I had one situation where the normal activity level was 150 and
one day, when there were too many runaway jobs, it went up to 650 - and
stayed there. This is usually not a major problem, though, because most
systems these days can handle the processing fairly if you set up and tune
using a good set of shared pools.

4) Without ANY doubt, 99% of performance problems are application related.
We recently spent several days tuning an iSeries to handle the workload from
several SQL servers. They were providing slow response at certain times of
the day. Even if we improved performance by 10% with a more efficient work
management configuration, the users did not notice the difference between 10
seconds and 9 seconds. Even if we had got 25% improvement, 7.5 seconds is
still a long time. It turned out to be a query that had not been written
well, had NO supporting indexes pre-built, and was running during these peak
times. By improving the performance of the query, we removed the performance
problem completely. Lesson learned - don't spend too much time on
performance configuration. Do it once, then review regularly, and prevent
programmers from changing things on the fly.

5) Increasing the length of a timeslice is the opposite of what you should
do... Oh.. .that is not related to pools :-)

- Trevor



----- Original Message ----- 
From: <rob@xxxxxxxxx>
To: <midrange-l@xxxxxxxxxxxx>
Sent: Monday, November 22, 2004 3:19 PM
Subject: Why separate pools?


> Since many people are setting QPFRADJ at 2 or 3 I think this begs the
> question:  "Why use separate memory pools?"  QPFRADJ is supposed to
> balance memory between pools.  However IBM puts some limits on it to make
> sure that not everything swings one way or the other too radically.  This
> causes a problem when you do have a major job shift and you do want a more
> radical shift.  If I am using QPFRADJ in this manner that why not just put
> everything into one or two pools?  I could still have numerous subsystems,
> (if that buys me anything), but why not run them all through the same
> pool(s)?