newer Linux nfs client freeze under load to OpenSolaris server

newer Linux nfs client freeze under load to OpenSolaris server

[Scott Serr]

My Ubuntu 10.04 workstation automounts my OpenSolaris 2008.11 server.  This works and is stable for "light use."  If I hit it hard though with large "cp -a" reads or rsync/unison, then Linux locks up.  The system is non-responsive, the activity light on my switch shows very active until I force the power off on the box.  The activity light to NFS Server is calm during the Linux NFS Client activity.

I believe this happened when I ran Ubuntu 9.04.  It does not happen on 2 other boxes: Ubuntu 8.04 and Gentoo ~2009.  These other boxes are on the same switch 1Gb as the NFS Server, the problemed 10.04 workstation has 2 1Gb switches between.

I've tried nfs mount options vers=3 and udp and tcp, none seem to solve the problem.

I put another 100M switch between my 2 1Gb switches, this seems to work around the problem - no crashing.

Ideas?  Seems harsh that Linux would freeze.
(I'll try with no switches next.)

Thanks for any ideas,
Scott

Re: newer Linux nfs client freeze under load to OpenSolaris server

[Trond Myklebust]

On Thu, 2010-07-15 at 10:44 -0600, Scott Serr wrote:
> My Ubuntu 10.04 workstation automounts my OpenSolaris 2008.11 server.  This works and is stable for "light use."  If I hit it hard though with large "cp -a" reads or rsync/unison, then Linux locks up.  The system is non-responsive, the activity light on my switch shows very active until I force the power off on the box.  The activity light to NFS Server is calm during the Linux NFS Client activity.
> 
> I believe this happened when I ran Ubuntu 9.04.  It does not happen on 2 other boxes: Ubuntu 8.04 and Gentoo ~2009.  These other boxes are on the same switch 1Gb as the NFS Server, the problemed 10.04 workstation has 2 1Gb switches between.
> 
> I've tried nfs mount options vers=3 and udp and tcp, none seem to solve the problem.
> 
> I put another 100M switch between my 2 1Gb switches, this seems to work around the problem - no crashing.
> 
> Ideas?  Seems harsh that Linux would freeze.
> (I'll try with no switches next.)

Does this bugzilla look relevant?

     https://bugzilla.kernel.org/show_bug.cgi?id=16056

If so, would you mind testing the patch that I posted for it?

Cheers
  Trond