[Samba] Connection reset by peer - HP-UX 10.20 & Samba 2.0.10

[Windsor Dave (AdW/MOE2.1)]

Hello All,

We have begun to experience a problem with our main production fileserver.
It is an HP 9000/D270 running HP-UX 10.20 and Samba 2.0.10.  I realize this
is an older version of Samba, but this system supports a 24x7 manufacturing
environment, and upgrade opportunities are not readily available.  In the
past several days, we have suddenly, several times per day, been losing
connections with our clients.  The clients are mostly NT, with a few Win2K
mixed in.  We get "waves" of problems where multiple clients are not able to
write to the server.  In my log.smb, I see repeated messages like

[2002/03/18 13:50:07, 1] smbd/service.c:(581)
  tester03 (10.11.12.13) closed connection to service tester_data
  read_socket_data: recv failure for 4. Error = Connection reset by peer

repeated for each client that was disconnected.

When we answer error messages like "Could not open file Y:\uptime.log" the
applications pick up where they left off and things are fine until the next
"wave".  There have been no changes to the server or network of which I am
aware.  Some of the clients have had some special application-specific I/O
cards installed recently, but this problem is also occurring on clients that
haven't been touched.  Any suggestions as to where I should start looking?
I don't have any real network analysis capability (I have an *old* HP
Network Analyzer, but it doesn't work well in the era of switched ports and
Gigabit Ethernet backbones).  I have Samba 2.2.3a running on a test server,
so I could upgrade if it is warranted and I can find a hole in our schedule.
I've been wanting to do so for some time now anyway...

Thanks in advance for any help!

Best Regards,

Dave Windsor
Robert Bosch Corporation
mailto:Dave.Windsor at us.bosch.com