We have a AlphaServer 800 5/333 acting as a RAID filestore server with
around 40-50 client alphastations mounting AdvFS via NFS.
Since we've moved up to a normal workload, the shell-level NFS performance
has dropped off markedly. For MH users, the delay to INC and SCAN mailboxes
can be obscene: 30-40 second delays to begin to show the results.
I suspect NFS or some kernel level issue is the problem. The diagnostic I
have so far is that a ls or any other command which causes a client to stat
the directory seems to show the delay more than anything else, apart from
MH's use of lockf() calls. Once there is a valid directory scan on the local
client, things seem to be much smoother.
Can any of the resident experts suggest to me ways to confirm there is an
NFS related problem, and any likely fixups? I found nfsstat didn't seem to
give much information.
All kernels are essentially un-tuned. We did increase the number of nfsd
threads to double the initial settings, to no avail.
cheers
-George
Received on Mon Feb 09 1998 - 23:56:47 NZDT