2008/8/1, Tor Arntsen wrote:

> - I have rebooted the computer in question and released some allocated
> memory and swap in the process. This has made more memory available
> for the system, and the installed gm4 is now able to go through
> without running out of memory. It appears to be eating up about 100MB
> each time it goes through. Before the reboot it would exhaust the
> available memory a bit before it got to that point (according to
> vmstat). (The system has 512MB RAM).

Aha!. It had to be something. At least now there is a real bone to chew at.

> - There was still apparently lots of swap space available even before
> the reboot, that didn't seem to avoid the resource problem though -
> Solaris may be behaving a bit funny in these cases I assume.

Memory manager returning no memory available to perl or gm4 while it
swaps out and obviously program not retrying on memory allocation
failure ???

> I'll do more testing at a later time, starting with a self-compiled gm4 1.4.5
> on the Solaris computer, and checking the memory usage compared to
> the blastwave version and other gm4 versions as well.

If it is an OS or OS configuration issue, the blastwave version should
be the best one as it is supposed to be Solaris specific and might
'know' its internals. But don't leave the perl binary out of the
equation, it might be the real culprit, after all aclocal is a perl
script which executes autom4te which is another perl script which in
turn executes m4.

At least now we're on the right track.

-- 
-=[Yang]=-