[SASAG] NFS Timeouts any Tips?

[Nick Webb]

On Thu, Jun 5, 2008 at 6:33 PM, Skylar Thompson <skylar at cs.earlham.edu> wrote:
> Nick Webb wrote:
>>
>> Just one client and one server, thus the crossover cable, but a decent
>> amount of traffic.  This looks warm:
>>
>> th 8 18266 8721.520 3868.030 826.070 0.000 311.490 138.490 101.280
>> 62.720 0.000 212.920
>>
>> The last (14th) column is high?  Doesn't sound horribly bad that we've
>> had about 4 minutes all day where were were nearly maxed out, but that
>> could certainly be an issue.  Looks like we are running 8 procs by
>> default:
>
> I think that's mostly OK, but it could be causing your problems if it pushes
> NFS from minor timeouts to major timeouts, where the interval between
> retries is much longer. In my experience, it doesn't hurt to increase the
> number of threads to some really high number if you're working on a
> dedicated NFS server. This will guarantee that the disks are going to be the
> bottleneck, and not the lack of NFS threads. They take up little RAM aside
> from the process table entries in the kernel.

Yeah, upping the number of processes didn't really seem to help much,
still got some timeouts.  I noticed that my remounts of /home with
"hard,intr" didn't really take (after a second look at /proc/mounts).
After I corrected that by unmounting and remounting /home, the errors
are gone, so far.

Thanks for all your help!  Now I know a lot more if I ever have to
troubleshoot some higher performance NFS setups.

Nick