TCPError 10048

[MBaas]

I have a 10.1-app-server running on a W2003-server (SP2) and for years it was running reliably. Yesterday the client's IT patched the server - and suddenly the app lost client-connections and reported TCPError 10048.
So they undid the patch, but for some reason things even "got worse" (as the client said) - it seems the time-interval in which the problem occurred came down to every 15mins (after restarting the app).

Has anyone else recently observed strange crashes after applying the latest MS-Patchwork? Any ideas/comments/suggestions?

[AndyS|Dyalog]

I wonder if you're getting abortive closes on the sockets ? In that case, the timeout period can be quite long before the kernel frees the address, so socket error 10048 is not unlikely.

It might be worth examining the possible timeout settings that can be set in the registry at both ends of the link. Unfortunately there's a goodly number of places in the registry where TCP-related values are set: two it may be worth searching for in MSDN are

HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\Tcpip
HKEY_LOCAL_MACHINE\System\CurrentControlSet\Services\VxD\MSTCP

These and other places may lead to the solution (and, I fear, potentially open a can of worms on what _could_ be configured !)

For anyone who wants to have a little more detail about the possible errors that TCP sockets can generate under Windows, then this is a good place to look: http://msdn.microsoft.com/en-us/library/ms740668%28VS.85%29.aspx.

[harsman]

10048 is "Address already in use" and is usually reported when a server tries to bind to a port that already has a listening socket due to some other process or connection.

Make sure that there isn't several processes listening on the same port (try "netstat -a -n -o" and grep for whatever port you usually listen on). If your application restarts itself, note that it can take a while for the port to become available, so it it restarts to quickly after shutting down, you might get this error.

We've also seen instances where software firewalls cause a different error message to be returned, e.g. some versions of McAffee turn 10061 Connection refused into 10060 Time out which can be confusing.