Tom,

regarding your server slowdowns and getmain/freemain,
what is the region parameter on the JCL to start the server?
I used to say '0m' and had problems like yours but not as severe.  I said '0m' 
because
that is what I do with db2 and if it is good for db2 I figured
it would be good for tsm. But IBM said that was bad so I
raised it to 512m and thing got a lot better.

I have increased it from there to 1792m.  TSM makes a cellpool
which is some percent of the regionsize.  It doesn't mean the
server is actually going to use 1792m of virtual or real.

I run a week without any problems.  I restart weekly because I am
afraid of memory leaks and then getting into the getmain/freemain
pit.

If you raise the region size, check with other syspgmrs to be sure it
will be honored and not overridden by some exit.

Hope this helps,

Bill Colwell

At 02:43 PM 4/8/2003, you wrote:
>> The server won't stop until the running process completes. Use an mvs cancel
>> command if you want the server down.
>
>That was what we ended up doing. We subsequently ran a batch job to
>update IDL definitions for the Web interface and then brought up the
>server without the /upgradedb parameter. The server behaved normally
>in both cases. When IBM finally responded to our severity 1 call they
>had nothing to offer except the suggestion that we should have taken
>a memory dump before cancelling.
>
>We are still seeing the same performance bug that we had under 4.2.3.2.
>Server performance gradually degrades. When the server approaches
>complete immobility Omegamon shows it spending nearly all of its time
>doing getmain and freemain operations, even though 'show memu SHORT'
>reports Freeheld well below MaxQuickFree. We have to recycle the
>server two or three times per day.
>
>As far as our site is concerned, all the talk about the most stable
>server release in years is just talk.

----------
Bill Colwell
C. S. Draper Lab
Cambridge Ma.