Hello,
I've a strange problem with a box I'm setting up as nfs-server
under 7-stable :
- tyan S2895 MB, 2*285Dualcore Opteron, 4G-ECC, ahd-scsi, nfe-network
- stripped GENERIC as kernel
- sources as of last saturday afternoon (European time)
I removed everything from /boot/loader.conf and /etc/sysctl.conf, still
I get "easily" data corruption when exporting ahd-scsi over nfs
(NB exporting geom_raid5 gives same data corruption)
Testing with the following pseudo code :
while checksum1 == checksum2 do
create random file of $1 MBytes
calculate md5 checksum1
copy
calculate md5 checksum2 on copy
Tested on both (as nfs-client) a 6-stable-i386 from a couple of weeks
ago as well as a linux 2.6.15-gentoo-r1 of about two years ago :
within half an hour the copy will be different .... ;(
I played with nfs-options on client side (nfs[23], conn, intr, [udp|tcp],
-r=, -w= ) but none seem to matter.
Start/Stop rpc.lock/sttatd on server/client just provoked some :
cp: utimes: BIG2: No such file or directory
cp: chown: BIG2: Stale NFS file handle
cp: chmod: BIG2: Stale NFS file handle
cp: chflags: BIG2: Operation not supported
cp: BIG2: Stale NFS file handle
cp: setting permissions for `BIG2': Stale NFS file handle
cp: closing `BIG2': Stale NFS file handle
[and then the while loop continued ... as if the NFS handle where not
that stale ..]
Anyway, I'll try to nail this down more (e.g. nfs-write performance
is horrible ... (nfsd falling down to 0% cpu and then after while
'wake up' and be at around 3-6% again))
I didn't stress-test this MB for a while, but last time I did was
with 7-PRELEASE/RC?/CANTremember-exactly-but-close-to-release
and all worked great
I did add 2G ECC to the 2nd CPU since, though I doubt that interferes
with NFS.
Bref, if anyone has a suggestion ???? (I will try downgrade
to RELENG_7_0 iff noone has a new suggestion for RELENG_7, but I'd like
to go forward and test some maybe suspect recent MFC or other
suggestion)
Thanx in advance,
best, Arno