[Beowulf] Fluent problems on a cluster

[Wayne Mallett]

G'day,

I am having problem with Fluent (probably the configuration of the 
cluster of 16 linux PCs with switched 100Mb/s connections) and 
wondering if anyone here can help.  I use the command fluent 2ddp -t# 
-pnet -cnf=hostlist to attempt to run Fluent on # machines within the 
cluster.  When #=1, everything is OK.  When #=2, Fluent only consumes 
about 20% of each CPU.  When #=4, Fluent only consumes about 5% of each 
CPU.  When #=8, Fluent only appears in a "top" display every couple of 
seconds and usually only registers as 0.00% of each CPU.

Some notes which might help:
Fluent doesn't appear to work at all with -pnmpi.
The cluster experiences ssh breakdowns from time to time, requiring a 
restart of the ssh daemon.
When MOSIX was running the systems crashed frequently, so it was 
removed.  There have been no crashes since MOSIX was removed.

Has anyone seen behaviour like this before, and can offer some 
assistance?

Regards,

Wayne Mallett

Dr. Wayne A Mallett
Phone [O]  -  0747815084
Phone [M]  -  0439875841
Email  -  Wayne.Mallett at jcu.edu.au
Smail  -  Information Technology & Resources
	James Cook University
	Townsville  Qld  4811
	Australia