[Beowulf] max number of NFS threads: NFS config optimization

[Chris Samuel]

----- "Rahul Nabar" <rpnabar at gmail.com> wrote:

> For a while I've been seeing errors of this sort in my
> /var/log/messages

Are you using ext3 for that filesystem by some chance ?

We long ago switched from RHEL to to Debian for our NFS
servers so we could use XFS rather than ext3 as ext3 just
couldn't keep up with the workload even back then.

My understanding from a talk at LCA on Linux scalability
many moons ago was that ext3 is single threaded through
the journal daemon, so our theory was that as soon as you
start to get a lot of writes to the filesystem they all
start backing up waiting for the their request to be
serviced. :-(

We've seen the same with Zimbra on ext3 filesystems,
but there we were able to ameliorate the problem by
extending the ext3 commit interval from 5 to 15 seconds
(it's just a mount option).

Could be worth a shot, and you can try it live by doing:

mount -o remount,commit=15 /home

You can set it back to the default with:

mount -o remount,commit=5 /home

YMMV, no warranties, batteries not included, if it
breaks you get to keep both parts.. ;-)

Best of luck!
Chris
-- 
Christopher Samuel - (03) 9925 4751 - Systems Manager
 The Victorian Partnership for Advanced Computing
 P.O. Box 201, Carlton South, VIC 3053, Australia
VPAC is a not-for-profit Registered Research Agency