Hi Eric,
On Jun 25, 2009, at 5:42 PM, Eric Anderson wrote:
Rush Manbert writes:
[ seeing TCP resets ]
Nice summary. :-)
You may be seeing the same thing we do under some stress tests with a
different app. The problem we tracked down to Linux re-using port
numbers before the timeout has expired. We saw something a little
different, namely syn, syn-ack, rst. We debugged it using wireshark,
which just reported that the ports were re-used. I also wrote a
script that looked at port-reuse time and found that it happend a lot
under the 60 second (might be 30) timeout, but it only rarely mattered
to us because we had many different servers.
syn, syn-ack, rst is pretty much what happens in the client/server
exchange that fails. I'll get wireshark and see what I can find out.
One good thing about this whole exercise is that I'm learning how to
use a bunch of new-to-me tools.
We found this on Linux 2.6 kernel. I'd recommend capturing a trace
and examining in wireshark. We tried lots of tcp options, but it
couldn't find one that had a useful effect.
-Eric
Thanks for the info!
- Rush
