Thanks to all who replied to this question. I described a situation where a
user
contends that he has a program that requires increasing amounts of memory as
it runs. When he runs it with sufficiently large parameter sets, it eventually
causes the system to hang, repeating "swap space below 10 percent free".
During the most recent hang, it killed off the inetd, while killing processes
in an attempt to free up swap. We are configured for "lazy swap".
Answers were as follows:
* it is possible to limit memory usage with a "limit memoryuse 100m" type
command. (not clear to me if this is swap plus mem or just mem)
* In "eager swap" mode, the offending job is killed. In "lazy swap" mode, the
system cannot necessarily determine the offending job and will kill whatever
process is resident when the situation becomes critical.
* Basic unix behavior is like "eager swap", where you allocate as much swap
as you need at process startup. "lazy swap" favors the process that needs the
space, to the detriment of other processes.
* adding more swap would help the problem.
* we could run a script that checked swap space in use, and before it became
critical kill the offending process.
Thanks for all the info. The user decided to check for internal ways (within
fortran) to find out the amount of swap space remaining - anyone know what
the call would be? - and try to kill his job internally before things become
critical.
Judith Reed
judith_at_npac.syr.edu
systems_at_npac.syr.edu
Received on Wed Jul 05 1995 - 19:33:23 NZST