You are seeing the effects of the Linux kernel Out Of Memory (OOM killer). When the kernel reaches low memory conditions, it takes decisive action and attempts to free memory for continued operation by looking for processes that are big consumers of memory. The algorithm takes into acount a number of factors, including age of the process, the time it last run, whether or not it is root and so on. This is merely a *symptom* of a system that is either badly overloaded or suffering from serious defects in memory.
You should also see information from the kernel at the time of the 'killed' message that indicate how much memory was available
at the time.
Recommendations would be to:
- update to the RHEL3 U8 kernel (2.4.21-47)
as there have been subtle adjustments to
the oom-killer and the vm
- consideration of using ulimit to limit
users virtual memory consumption
- limiting high io activity such as backups
to slack periods of time
- adding more memory to the system.
- use of top and ps to get more of a picture
of the application use
- use of sar data to get a picture on overall
memory consumption
- monitoring /proc/meminfo will give you some
more granularity on where the allocations
are occurring.
Necessary questions: Why? What? How? When?
