Matt, I had a similar problem and called support; level 2 explained
why region=0M is bad.  TSM takes the JCL region size and getmains
a small fraction of that to use for a cellpool.  The object of the cellpool
is to avoid doing getmains and freemains.  When it sees 0M a
too small default cellpool gets built.  I changed my JCL from 0M to
1280M and the problem went away.

When you see your server take all the CPU and stay there, it has
probably run out of cellpool space and started doing getmains.
I suspect there is a bug since cancelling processes doesn't always
make the server return to its normal low CPU state, but I haven't pressed
for a fix since changing the JCL avoids the problem.

Hope this helps,

Bill

At 07:57 AM 7/17/2002, you wrote:
>Hello all,
>        I am looking at the TSM 5.1 Quick Start book for Os/390 + z/OS.  In
>chapter 2 there is a section  'REGION SIZE FOR YOUR TIVOLI STORAGE MANAGER
>SERVER.  The section goes through a process for determining the best region
>size to assign to TSM.  The last sentence says,  'Specifying 0M will result
>in poor server performance.'    WHY?
>        I am dealing with a long outstanding, not solved problem.  I have my
>region size up at 1000M.  (I am using a 9672-x57 that has about 7GB of
>memory on the lparand a TSM server with a 25GB DB and 230 clients).   The
>problem is TSM will 1 or 2 times a week  start using all of 1 processor to
>do some simple work, like run 5 migration processes.   I will take TSM down
>and up.  It then continues on getting work done faster and using 1/4 of one
>processor.   Does TSM do something with memory assignments internally that
>too large of a REGION causes a problem?
>Matt

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