NFS server not responding

[Alex Lavrov.]

Hello,

I have an nfs filesystem mounted on linux RHEL3 from HPUX.

Every 4-5 minutes I get in /var/log/messages this message:
kernel: nfs: server serv1 not responding, timed out

in "nfsstat -c" I see that:
calls retrans
708028 77304

It's not so good. What are the steps that I can take to solve this problem?

The network cards on both servers configured on 1000, Full, Autoneg ON.

No problems in pign either.


Thanx.

[Alex Lavrov.]

Another thing, maybe it's related, I also get a lot of these messages:

kernel: application bug: dw.sapWT3_D00(3602) has SIGCHLD set to SIG_IGN but calls wait()

[Aris Cruz]

Oh joy, nfs problems, they are always fun...

couple of things you can try.

1. change rsize=8192, wsize=8192 in mount options

2. RHEL3, i believe still has a default mount of udp, you can change the mount to tcp and nfsver=3

[Alex Lavrov.]

Hello,

I'll try the first option.

About the second, it was TCP, so when the problems started we changed it to UDP it didn't help and it's already version 3.


thanx.

[Dave Falloon]

watch out for soft mounts. The linux NFS faq ( http://nfs.sourceforge.net/ ) covers this under the heading at the bottom:

Why do I get NFS timeouts when I mount a Linux NFS server from my Solaris NFS client?

This applies to linux clients as well. These messages are shown when a client gives up trying to connect the the nfs server, this only happens with soft mounts, a hard mount has no timeout value. Two settings to mitigate this are "-o proto=tcp,hard" Since you are using Gig E you should also adjust the performance options.

--Dave

[Alex Lavrov.]

Hmm, good point. I use soft mounts.

Before I assign points, I'll check it :)

[Dave Falloon]

Please do as there could be other reasons for timeouts such as process limits on the server, sometimes you can run low on filehandles and start bouncing connections, etc.

--Dave

[rick jones]

You might start looking at switch stats, ethtool stats (linux side) and lanadmin stats on the UX side. Also, netstat -s on linux and netstat -p udp and netstat -p tcp (depending on the flavor of the mount) on the UX side. Might toss-in a netstat -p ip if using UDP.

The idea is to see where packets are being lost and leading to the retransmitted NFS requests.

I don't have one for linux yet, but

ftp://ftp.cup.hp.com/dist/networking/briefs/annotated_netstat.txt

may be of some help.

[Steven E. Protter]

Is the physical networking solid on this box?


I would suggest turning Autonegotiation to off and see if it gets any better. I would also see if other boxes have NFS problems with this box, perhaps even an HP-9000 or Solaris box.

The client and server nfs software should be up2date or yum current.

SEP

[Ivan Ferreira]

We had connection problems with an ethernet adapter on linux, and we disabled the apic in the kernel and the problem was solved.

linux noapic

I also agree that the autonegotiation should be turned off, and configure fixed speed on both, the switch and the server.

[rick jones]

In the original post, the speed was said to be "1000" which I would take to mean gigabit. In that case, autoneg _must_ be on - gigabit requires it.