Posted by Jim Guerber on November 19, 2007 at 14:27:27:
Well, I think we have this problem solved (finally!).
We opened a case with Microsoft on this issue. They supplied a "sniffer" program to monitor the network traffic and we can see what went on while a disconnect was happening. We still do not know what change in 2003 server sp2 caused the problem. I have a list of changes from Microsoft, but nothing clearly spells out what went wrong.
We got in touch with the author of the framework code that CfamNetServe uses. Here is what he said:
"I've spent a couple of hours looking at the code this evening. I'd quite
like you to finish switching to using the latest version of the code (which,
I think, Brian is in the process of doing now) before I spend any more time
looking at your code as this could fix your problems.
The issues that you're experiencing could be due to how the old version of
the framework dealt with connection errors coming out of the IOCP. For the
errors that were assumed to be fatal (which, in that version, were too
numerous) the exception thrown would destroy an I/O pool thread and that
would cause outstanding I/O requests that were issued by that thread (on any
connection) to be aborted. The latest code deals with these errors
better and could potentially fix your problems. As for why you're suddenly
having these problems on some machines, I think it would be better for you
to try the latest version and then if that still gives problems I'll
take another, more detailed, look."
Brian did indeed switch over to the new framework code and this solved the problem.
I was about to announce this release on Friday when our test account called with another case of error 61/9b. I think we traced this to the user's machine going into standby mode. This is important for all of you. Our network code periodically sends packets BACK to the workstation even when nothing is happening on the workstation. If one of these packets does not get through, the connection is destroyed. In this user's case, Standby caused her network card to no longer function. We are discussing how to adjust our code to this, but for now, make sure the workstation does not go into standby or hibernate.
Server version 705 contains the fix to this problem. I urge you to install it at your earliest convenience.
Many thanks to all of you who participated in this solution.