In some email I received from Scott L. Burson, sie wrote:
[...]
> Oh -- I guess I left out a key piece of information: we use large, long-
> running CASE applications here that build up large amounts of state within the
> process (250MB process sizes are not unusual around here).  It can be a major
> hassle to rebuild that state after logging out and logging back in.  This is
> exactly what we are trying to avoid.
> 
> 			    TCP connections shouldn't survive 24 hours of
>    interruption in network connectivity.  In fact, I believe there are
>    standards that would be violated all over the place by allowing this
>    sort of thing.  Perhaps you should get the TCP RFC and read it?
> 
> Maybe I should, but I don't really see what harm is done if we choose to
> violate some such standard internally.

You need to turn off "keepalives", and this is usually turned _on_ by
applications, although I could be wrong and the default is "on".

Either way, turning it off in your apps. should stop this, so long as
there is no network traffic during the interruption.  If you do have
packets going back and forward, and there is a break, it makes sense for
it to drop out.

darren