[Tfug] scp transfers stalling

[John Gruenenfelder]

Hello all,

I've got a problem with scp file transfers stalling and never recovering.
It's happening on only one of three Debian systems I use, though, and I can't
seem to figure out the cause.

The machine with the problem is running Debian/testing using openssh
4.7p1-12.  For the working machines, one is Debian/unstable with 5.1p1-2 and
the other is Debian/stabe with whatever version of openssh it currently has
(it has the Debian key-bug fixes, but I can't access it right now).

The stalls seem to occur only in one direction: when I copy data *from* the
machine to mine over the Internet.  It does not seem to happen when I scp
files to the machine.  The stalls usually, but not always, occur early in the
transfer of a particular file.  The stalls are not constant, though.  If you
repeat your scp command again and again, eventually it will work.  Stalls seem
to happen roughly 50% of the time.  I really wish scp could resume
transfers...

The main difference between these three machines in their Net connectivity.
The stalling machine is at the UA and therefore has loads of bandwidth.  For
the working machines, one is my cable broadband with *very* low upload speed
(no more than 320 kbit) and the other is on a DakotaCom DSL link with
reasonable upload speed (still nowhere near UA speeds).

I'm not doing anything tricky with the network or SSH settings; it's all
pretty standard.  My first thought was that it was the poor router the machine
plugged into.  There were other problems with it and I was hoping this was
another one of them.  However, the router has been replaced by a nice WRT54GL
(though not yet with non-Linksys firmware) and while it fixed the other
issues, it did not fix the stalling.

It's also not the link from UA to me or the network on my end as this same
stalling problem is happening to other users connecting from varied networks.

I'm running out of ideas and things to test.  Having scp be more verbose gives
little extra output and shows nothing useful in regards to this issue.  At
first the problem seemed to be linked to the Debian key-bug fixes, but further
checking revealed that it actually began shortly before that problem was
fixed.

Other protocols seem to have no problems.  For example, if I'm having too much
trouble with scp, I can just stuff the file into ~/public_html and grab it
with wget and have no issues at all.

Possibly related to this scp stalling is that my SSH sessions to this machine
occasionally stop responding and never come back.  They need to be axed with
the tilde-period combo.  Especially annoying when you're typing an email.
This problem, though, occurs much less frequently than the scp stalls and
usually only happens after I've let an SSH session sit idle for a while.
Perhaps it has something to do with data volume?

Does anybody have ideas as to what might be causing this?  My boss really
wants some answers and I need to give them to him.  :)


-- 
--John Gruenenfelder    Research Assistant, UMass Amherst student
                        Systems Manager, MKS Imaging Technology, LLC.
Try Weasel Reader for PalmOS  --  http://weaselreader.org
"This is the most fun I've had without being drenched in the blood
of my enemies!"
        --Sam of Sam & Max