Scenario:
Server: AIX 4.2.1.0; ADSM 3.1.15
(No clients involved - server issue)

We had a problem yesterday where our ADSM server kept crashing. Called IBM
Support, and they were concerned that my /usr directory didn't have enough
free space.

Even before we called IBM, we noticed that ADSM had created a 50Mb+ core
dump in /usr/lpp/adsmserv/bin, and that /usr was 100% full. Deleting the
core dump, or course, freed up over 50 Mb and  reduced /usr to about 94%
full, but ADSM continued crashing, after running a few minutes each time it
was started, either in foreground or background. In each case, it was
reporting a ANR7834S (segmentation error).

The IBM Support person had me extend the volume group for /usr, and if I
follow the numbers it was increased by 32Mb, and /usr utilization dropped
to 91%. It then stayed up, and our overnight backup series ran cleanly as
far as I can see.

Again, if I follow what's happening here, our DB and log are on different
logical volumes (and physical spindles), so that shouldn't be affecting
/usr. Of course, there's plenty of stuff unrelated to ADSM in /usr on this
machine.

Finally, my question: Is there documented officially, or any rules of
thumb, as to how much free space one needs to maintain in the filesystem
where the server is running? From my unplanned empirical research
yesterday, it would appear to require more than 50 Mb but less than 80Mb.
The IBM Support person said he preferred /usr to be below 80%, but my
question is, "80% of what?". The filesystem /usr isn't a forced, standard
size in AIX, is it?

The whole idea of the ADSM crashing and leaving a core dump in
/usr/lpp/adsmserv/bin and /usr being 100% full would seem to be a "chicken
and egg" question - Which came first?

Any insight would be greatly appreciated.

Regards, Steven

Steven J. Chaba, Lead Analyst
Distributed Systems Technical Support
Rochester Gas & Electric Corp, Rochester, NY
716.771.2137 voice