At 22:59 Uhr +0100 28.11.2007, Manuel Bouyer wrote:
>>  NFS over UDP works very badly if the NFS server is faster than the client
>>  or has faster networking. My NetBSD-i386 NFS server (1Gb/s ethernet)
>>  managed to make my SPARCstation LX NFS client (10Mb/s ethernet) very
>>  unhappy until I switched to NFS over TCP.
>
>In my case, the NFS server is much slower than the client, and I suspect
>it's the condition to reproduce it.

TCP mounts here, clients and server are of the same machine class 
(i386 P4 / AMD Athlon, all netbsd-3), and I still see the occasional 
stalls as well as the "...not responding / is alive again" noise:

[hf@Sagran] ~ > w
10:16AM  up 12 days, 16:35, 2 users, load averages: 0.55, 0.37, 0.33
USER     TTY     FROM                        LOGIN@  IDLE WHAT
sarmadi  :0      -                          19Nov07     ? /bin/sh /usr/pkg/bin
hf       pts/2   dachstein.nt.e-technik.tu- 10:15AM     0 w
[hf@Sagran] ~ > dmesg | grep -E "(not responding|is alive)" | wc -l
       50
[hf@Sagran] ~ >

i386 Debian 4 clients exhibit stalls too, so I'd suspect the NetBSD 
nfs server implementation as the common cause...

	hauke


-- 
      The ASCII Ribbon Campaign                    Hauke Fath
()     No HTML/RTF in email            Institut für Nachrichtentechnik
/\     No Word docs in email                     TU Darmstadt
      Respect for open standards              Ruf +49-6151-16-3281