Process taking over system

[Jonathan Eisenhamer]

Versions: SunOS 4.0.1 on a sun network consisting of a 3/260 as NFS/YP
server and four clients, all diskless consisting of one 3/160 and three
3/50's.

The problem:  When a compute-intensive process starts running and the
3/260 server, the NFS response to the clients drops to the point of being
unusable.  All the clients start a cycle of losing the NFS
server/regaining it.  The result of this is that logging onto clients can
take up to 15 minutes, and trying to run sunview or X is not practical at
all.  However, usage on the server itself is not significantly affected
(in fact most users on the server don't realize that there is a problem).

What has been tried:  The only "different" hardware used by such processes
is usually the Sun 3 fpa board.  However, recompiling the programs not to
use this has not helped the situation.  Also, I have tried renicing such
processes to 20 with no effect.  I have tried renicing the nfs deamons
(biod, nfsd, yp*) to negative values (up to -10) with no effect.  This
behavior is not seen with every compute intensive program:  Only one
program does it consistently.  Others seem to do it (but none run long
enough to check), and any cooked-up programs don't exhibit this behavior
at all.  None of these programs use much memory (the server has 32Mb
available and all the virtual memory sizes are under 700 pages according
to ps(1)), and very few disk accesses.

So what should I be looking at?  Are there parameters to the NFS that can
be changed?  What other resources are there that could be exhausted?  Has
this behavior been seen before?

Thank you for your time,

Jonathan Eisenhamer
jon at astro.ucla.edu
jon at uclastro.bitnet
bonnie::jon (span 5.708)
(213)206-8596