[Beowulf] SLES 10 and NFS Performance

[Joe Landman]

Tim Moore wrote:
> Hello All!
> 
> I have a small two-node cluster with AMD64 Opteron processors.  It was 
> running Centos 4 and NFS performance was good while not hindering 
> performance.  I replaced Centos with SLES 10 about 2 weeks ago and just 
> now got around to really hitting it hard submitting jobs.  Processes 
> that normally took about 15 seconds such as normal IO (each processor 
> writes its own data file) now require 10-15 minutes.  When I observed 
> the IO status, I saw that I could probably write the file by hand faster 
> than the 2nd node was writing the files to the server.  I would have to 
> check but I think the mobo has integrated broadcom ethernet.  I have 
> tried the usual tweaks with respect to rsize and wsize and performance 
> seemed to worsen if it changed at all.
> 
> Has anyone experienced similar issues?  And/Or does anyone have 
> suggestions outside the wsize and rsize?

Yes.  Disable jumbo packets, use TCP rather than UDP for mounts.  RHEL 
and variant clients had lots of trouble with NFS IO performance using 
broadcom chipsets talking to SUSE 9/10 in the past.  I filed a bug with 
RH many moons ago on it, and it got kicked because we didn't have a 
support contract for the machine.

Our workaround was to turn off jumbo packets on the client and servers 
and turn off UDP mounting from the client side.  This worked quite well 
in production at several customers.

Joe

> 
> Thanks!
> Tim
> 
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf


-- 

Joseph Landman, Ph.D
Founder and CEO
Scalable Informatics LLC,
email: landman at scalableinformatics.com
web  : http://www.scalableinformatics.com
phone: +1 734 786 8423
fax  : +1 734 786 8452 or +1 866 888 3112
cell : +1 734 612 4615