> The minimum and default (since z/OS 1.10) of 1M per CPU for
system trace is often adequate, and that is what is used in many
of the dumps I see.


As said, we currently have 15MB per processor, which seems to be luxurious I 
conclulde from the responses I got.


>The larger the trace, the more CPU time it takes to copy the data
into a snapshot.  Currently, most entries to RTM2 result in a
snapshot, before we know if any dump is going to be taken.


Ahh, this is very intersting! I had no clue about this. Thanks Jim. I'm not so 
much concerned about the (short term) fixed storage the snapshots require, but 
more about the CPU time it takes to copy this. This is an argument to make 
trace buffers not arbitarily large.


On the other hand, on our productions LPARs the 15MB cover only 2 seconds for 
the vertical high CPs. For the VM and VL CPs, the space is sufficient for hours.


The reason for all of this is I had to chase a program problem in produciton, 
not reproducible in test. The dump was taken out of Smart/Restart recovering 
for a follow-on error of an earlier PGM check 0A. I would have loved to see the 
trace entries for the 0A, but it was already overwritten, since the 0A and the 
dump lay more than 2 seconds apart.


--
Peter Hunkeler



----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN

Reply via email to