The problem appears to be in the way we were doing our network
configuration on the 4100. We have both Ethernet (10Mbs) and
FDDI cards, and for historical reasons (probably) were assigning
the system name to tu0. By trial and error, we found that if
we completely disabled tu0, we had no problems. This was most
likely the result of the NFS server only having a FDDI link,
and our new 4100 had both of its IP addresses on the same subnet.
One anomaly that this helps explain was why we would lose our FDDI
connection when we pulled the tu0 cable out of the switch. Why
that occurs is really still a mystery in technical terms, but as
a practical matter it is no longer an issue, since the solution
to all of the problems encountered was to identify the FDDI IP
address with the machine name, after which we have restarted the
Ethernet to see if it would come up. It did, but we're not at
the point where we feel comfortable advertising it to the other
servers.
There were two messages received Friday which got clobbered over
the weekend, but my thanks to those who responded, one of which
is quoted further below for information value.
Michael Watson
Unix SA, Dataline, Inc.
Virginia Beach, VA
ORIGINAL MESSAGE:
> >
> > Managers,
> >
> > We were wondering if there are any new developments on the NFS
> > mounts problems listed in the archives. We've got a new 4100/
> > DU 4.0D/patchkit #1/Firmware 5.1 which is trying to mount a
> > subdirectory from a similarly configured 4000. It appears to
> > do the mount OK, but as soon as we try to do an 'ls' on the
> > mounted directory, our network freezes. It usually stays
> > hung until we do a umount -f (or the network folks pull our
> > FDDI link) or reboot the system.
> >
> > The 4100 has both a FDDI and 10Mbs Ethernet running through
> > a switch. We've tried mounting the directory from the client
> > side with "-o nfsv2,soft,intr" and by the dbx patch to /vmunix
> > "patch do_client_readdirplus=0" and rebooting. Neither has
> > been any help.
> >
> > So, we're guessing the NFS through the switch might be the
> > problem, and the network guys are off-site right now. One
> > question we haven't been able to figure out is how we check
> > to see (if it matters, that is) if our interfaces are set
> > to half-duplex or not. We've tried ifconfig, netstat and
> > pfstat, but don't see anything helpful or didn't get the
> > right syntax (unless SIMPLEX in ifconfig is useful?).
> >
RESPONSE:
David Ritch wrote:
>
> Check the MTU.
>
> The symptoms you describe are similar to some problems that we have had
> using Sun FDDI's, DEC FDDI's and ATM's, and Fore Systems PowerHubs. The
> DEC FDDI interfaces support packets up to 4kB, as one would expect from
> FDDI. The Sun FDDI's only support ethernet-size packets, and silently
> drop any larger ones. As a result, we could nfs-mount files from our
> Sun servers on our DECs across FDDI rings, but could not ls a large
> directory, or load a large file.
>
> I only partially remember the problem with the PowerHubs. With an old
> version of their software, they were splitting IP packets from the ATM
> into 1500B packets. An upgrade allowed them to break them into 4k
> packets, as they should, and this broke our connection between DEC and
> Sun.
>
> It is safer to lower your MTU to 1500, but it may hurt your performance.
>
Received on Mon Jul 13 1998 - 19:35:41 NZST