Hi,

Here is a good one. I know that the 54 errors are caused by the client
not getting back to the right IP address on a server in this case. Here
is what we have done: we added a GigE connection to a media server that
has been being swamped. We left the 100bT IP and DNS entry alone, but
made the GigE the default route to the media server. The DNS is set up
to be round-robin.

All of our other media servers and our master server all are
multi-homed, with DNS round robin. They all work fine. Now, with the new
system, however, when we added the new IP and GigE many clients (not all
by a long stretch) fail with 54 errors.

The clients that are continuing to fail with 54 errors are mostly MacOS
9 and Windows < W2k (e.g. NT), though a few Solaris 2.5 systems are in
there, too. A reboot of the clients with the 54s fixes many, but not
all. Looks like some wierd DNS thing, really, but the DNS all resolves
fine, including from boxen half-way around the world (I tried a box in
India that I have a login on and that worked fine with nslookup).

You can see this yourself. nslookup flashback.lbl.gov every other time
will get the other IP address. This works fine with the master and other
media hosts, as I say.

We are running NBU 4.5 and have all the latest patches taht I am aware
of.

Anyone seen this before or have other thoughts? I have trolled through
the emails and have seen other posts about this. Nothing has really
helped, though I think it is DNS caching on the clinets, personally,
since a reboot (cold-reboot, mind you) seems to solve most of the
clients' problems.

I'll post more details as I find them....


Thanks in advance!

Chris