telnet disconnect problem

This is a problem that has been plaguing me, and I cannot figure out
what's wrong.

I have a Red Hat 7.3 server sitting behind a VPN. It's a fast machine.
I have client systems on the other end of the VPN using telnet to
connect. Telnet on our end is handled via xinetd.

Recently (past 3 weeks) they have started getting disconnects. After
certain amounts of inactivity they will get booted. so for example, if
they start something that takes a while, and then go to lunch, they'll
get disconnected, and the job will die.

The other end is currently blaming us. We both claim to have changed
nothing (it's true, at least for my end; I didn't change anything, and
the software didn't update itself..) but the problem persists.

I've done googling, and found two ways of addressing this, at least from
the server side:
1. Change the tcp timeout on the server to be higher
2. Change the xinetd.conf file for telnet to have KEEPALIVE as one of
the server flags.

There's no way I'm changing the global tcp timeout on the server via the
/proc interface - I don't want to afffect anything else - and I did make
the xinetd change.

The problem is persisting, and in the kernel messages, I see things like:
ttloop: read: Connection reset by peer
ttloop: peer died: EOF

The PID numbers don't jive with both types of messages (meaning that I
won't see pid 1100 show up for a Connection reset by peer message as
well as a peer died message. They will get one error, or the other, but
not both.) By tracking PID numbers, I have been able to find that they
aren't the only group of people getting those disconnect messages - of
course, I have no idea if Bob from whatever different VPN endpoint
disconnected by closing his telnet session properly, or just shut down
his computer directly, or yanked his ethernet cable.

I'm led to believe that because of the error messages, that it's not a
problem of our server kicking - it's a problem of their end
disconnecting, or dying. I have no complaints about VPN problems - it's
a small number of people getting disconnected in a larger number of
people. I'm trying to figure out where these people are in relation to
one another, to see if it's a problem in a specific office, but I don't
know, apart from sitting around watching tcpdump, how to figure out
exactly what's causing the disconnects.

Whaddya think?

Adam
---------
To unsubscribe, send email to <aklug-request@aklug.org>
with 'unsubscribe' in the message body.
Received on Fri May 12 09:21:43 2006