Kernel Memory Allocation and Swapping (long)

[Doug Toppin]

We are running IBM Xenix 2 on the 286 with 3-meg of memory
and have the following problem:
One of our processors has to dedicate 1.7 meg of memory to a
memory resident database. There are still about 20 programs
to be run all of which are blocked on the network or on message
queues. The dedicated database locks itself in memory with 'plock'.
Once the other applications are running, any command entered at the
shell produces only another shell prompt, the program is not run
and no error message appears. All of the processes already running
exhibit no unusual behavior, they still work correctly. After about
5-minutes the processor panics or goes away (no longer any keyboard
response). The kernel is about 350K which
leaves approx 1-meg of core for processes. It appears that memory
is being overutilized and swapping is not occurring properly. The
remaining applications total about 1.5 meg so there should be enough
memory to swap in and out as needed.
Has anyone seen similar behavior before?
If anyone out there that has the source to the shell could take a look
at it and see what it does when a fork or exec fails due to lack of
memory and then let me know I would appreciate it.
Also, what happens when a fork needs to occur and there is not
enough memory?
Can fork communicate with the swapper and get more memory?
We do not have the option of switching to SCO. We are stuck with IBM.
In general, if anyone has ideas on how to localize the problem
I'd like to hear from them.


thanks
Doug Toppin
uunet!melpar!toppin