Hi,
I had replies from:
O'Brien, Pat [pobrien_at_mitidata.com]
I would first begin by double checking the network adapters are
runing full
duplex AND corresponding hub/switches are MATCHED to the network
adapters.
I have several systems with nfs mounts and gb of copies happening
all the
time. Secondly I would remove the home directory from the
automounter, and
just update fstab to mount fs automatically on reboot.
John McNulty [john_at_jmtl.com]
Hmm .. NFS is normally quite robust. Check out the syslog log files
for other
errors. Also, is it just NFS or do you also get performance
problems with FTP?
If FTP is horrendously slow between these two boxes, then you have a
duplex
mismatch problem (half duplex talking to full duplex). Either
between the
interfaces of the two boxes, or a box and a router in between. Slow
FTP is a
dead ringer for this. I imagine NFS wouldn't like it very much
either.
Cohen, Andy [Andy.Cohen_at_cognex.com]
I really don't know the difference but I've 'heard' that NFS3 is
better than
2. I would also recommend you issue the NFS mount with the 'tcp'
option.
My recollection is that NFS uses 'udp' by default and I've been told
you'll
get better connections and throughput with tcp. Check the NFS man
pages for
more details.
in our /etc/fstab file we have nfs mounts like this:
remote_server:/mnt1 /local_mnt nfs rw,tcp,hard,intr 0 0
Fletcher, Joe [joe.fletcher_at_Metapack.com]
I had problems with nfs3 over tcp on 5.1. I fixed everything to run
over udp
and it started behaving again.
system administration account [sysadmin_at_astro.su.se]
NFSv3 is a lot better than v2, yes. Server-side, Tru64 supports
both.
Client-side, you can force one or the other with mount options ("man
mount"
for details). You may find it convenient to include the appropriate
option
in your automounter maps.
Moving to static NFS mounts
is likely to make things even worse (because a larger number of NFS
mounts
will be active at any given time). Yes, a brief delay is expected
when one
logs in and the home directory needs to be automounted; but in a
properly
tuned network this is short (maybe 10 seconds) and does not give
rise to any
"NFS server not responding" messages. I would rather tune the
network than
give up on the automounter. (We're automounting all home directories
and
many other disks without any of the problems you describe.)
Maybe you should try using tcpdump to capture the NFS traffic when
this
happens. That could give you a clue as to what is going wrong.
Well I'm not sure I got to the bottom of the problem. The little more about
the hardware situation
Machine: NIC Setting at prom router port
AlphaStation 200 4/233 10mbps twisted pair 10mbps
AlphaStation XP1000 2x10/100mbps auto-negotiate 10/100mbps
I tried setting the 4/233 to "twisted pair, full duplex", and although it
appeared to work for small packets, sending a largish file via ftp caused
the thing to clag. When I reset the NIC to "twisted pair" and reset the
network software on the 4/233, the ftp transfer worked quite well. So at the
moment all is as it was. It would appear that nfs can negotiate what level
of nfs is used. When the 4/233 was rebooted with the NIC set to "twisted
pair" the nfs service between the machines seemed to be at nfs3. I haven't
tried setting the tcp option. There seem to be arguments for and against on
that one.
Dennis
Original question:
I have 2 machines one an XP1000 running Tru64 V5.0A and another a
AlphaStation 200 4/233 running Tru64 V5.1. I have not applied any patches to
either system. The problem I am experiencing is that someone starts up a
multiple file copy, usually from the XP to the 200 over nfs, and then it
would appeat that things start to clag up. If I try to logon to the XP using
rlogin from my PC, it usally puts up the motd and then appears to get stuck.
After a little while it puts up a message saying something like "Waiting for
nfs2 server on the XP". The eventually the logon completes and at the logon
screen the machine appears to be travelling quite comfortably (ie there is
no wait while it does a "ps -ef", or some other application). However if I
logout and attempt to logon then the same wait after the motd re-occurs. So
then from /sbin/init.d I did an "nfsmount stop; nfs stop; nfs start;
nfsmount start", and what do you know I can then logout and log back in
without the wait.
I should add that automount managers the connections on both machines. All
mounts are listed in a /etc/auto.direct file.
What to do??.
(a) Well I was wondering about nfs3, what is it, is it better than nfs2, how
do I force nfs3 connections.
(b) upgrade the XP1000 to Tru64 V5.1 (can't go to V5.1A, because I don't
have a work around for the fact that ERDAS/Imagine8.4 does not work on
V5.1A)
(c) apply patches.
(d) get rid of automount and use nfs.
Thanks for any suggestions.
######################################
Dennis Macdonell
Systems Administrator
National Mapping Division, Geoscience Australia
mail: PO Box 2, Belconnen, ACT 2617
email: mcdonell_at_auslig.gov.au
ph: 61 2 6201 4326
fax: 61 2 6201 4377
######################################
Received on Thu Mar 28 2002 - 00:38:02 NZST