NFS - IRIX client issues

Hello -

I was wondering if there were and know NFS issues with IRIX clients?
I'm seeing a problem where an IRIX 6.5.17m client
accessing a linux 2.4.18 (redhat 7.2: -18.7.x) server will hang trying to access a mount. No traffic appears to make it to the server so it appears to be locked up on the client end, but I don't know why.

Thanks!

- Orion

Can you give a few more details? Does the nfs share mount, or hang at
mounting? What type of mount are you trying to do? (IE nfs v3 tcp, or v2 udp)

PS- Are you certain you aren't running iptables or ipchains? "/etc/init.d/ipchains stop" Also try mounting with out locking. "mount foo:/foobar /mnt/nfs -onolock" Have you tried nfs v2? "mount -o nfsvers=2 foo:/foobar /mnt/nfs"

The mount comes up fine and works for quite a while and then crashes. This is under relatively heavy load (tar files being unpacked, data files manipulated, etc.). No iptables/chains.

The mount is automounted, the resulting mtab entry on IRIX is:

lego:/export/turb3 /data/turb3 nfs vers=3,rw,dev=100007 0 0

I believe the mount is UDP, I'm not specifying any special options.

I'll look into trying nolock and v2. SHould I try TCP?

- Orion

Are you using amd or autofs? Does it occur when you manually mount
the share? "mount lego:/export/turb3 /data/turb3"

You could, but I was wondering if you could be hitting a TCP bug.

autofs. Do you realy think there is a difference between using mount and autofs to establish the mount?

Hm. IRIX doesn't seem to have a nolock option.

Followup: We changed the nfs mount options to mount with NFSv2 and have not been able to reproduce the problem, yet.

I'm having the same problem, with Solaris 8 on SPARC for the NFS server (as opposed to Linux), on one of my machines. For some reason it only happens when it's plugged into a 100MBps Netgear non-switching (i.e, "old fashioned" in a sense -- half-duplex) hub. If I plug it straight into the wall at work (this is connected directly to a 10MBps (I know), full-duplex (I think) port on some kind of switch whose other details I have no idea about), the problem instantly disappears.

At least, I think it's the same problem. When your connection collapses, does IRIX complain about timeouts trying to contact the NFS server, almost as if the NFS server fell off the face of the planet?

I just noticed this patch (4808: "NFS3 hangs with delayed writes, panics with imon") for IRIX 6.5.17m. The following page has more detail (although just reading the following page, as well as downloading the actual patch, requires an SGI support contract or warranty -- and I can't look at it and summarize it because TPTB at work have canceled the SGI support contracts with the intent of eventually replacing all the SGI boxes with Linux-based x86's or the like):
http://support.sgi.com/colls/patches/docs/browse/support/pinfo/pinfo4808.html

So, that patch might help, if you have access to it. I also found another document, "Pipeline: [Oct-Dec 2002] IRIX 6.5.17 NFS Changes and Tuning", which also cannot be accessed without a support contract or warranty:
http://support.sgi.com/search/?cmd=getdoc&db=pipeline&locale=C&coll=0650&highlight=type,pipeline,PipelineYear,2002,PipelineIssue,OctoberDecember&fname=content/pipeline/html/20020402NFS.html

Finally, I have no idea if IRIX 6.5.18m fixes any NFS bugs. If it does, it can be obtained with an M Series Access contract that costs $500 per workstation per year (servers not eligible, that is, they can only get 6.5.18m through support contracts AFAIK). A tiny bit more info on that here:
http://support.sgi.com/news/support/IRIX_M_Stream_Implementation.html
http://support.sgi.com/news/support/IRIX_M_Stream_Implementation1.html

I hope this helps. It might not help as much as other suggestions (such as trying NFSv2), but it might be better than nothing (especially if the machine is still covered by an SGI support contract or warranty).

-Barry K. Nathan <[email protected]>

Actually, I get exactly zero in the logs and aparently zero NFS traffic arriving at the server, so we may have different problems.

Thanks for the support links, but unfortunately I'm basically in the same situation - support cancelled in the belief we will eventually move completely to linux (though for now the SGI is our only large memory 64-bit platform).

- Orion