> > Having just browsed ISS-test4, do I read it right that there was a
> > generic memory leak in a _load_ of ethernet card drivers whenever a
> > transmit timeout occurred?? I have plenty of transmit timeouts in
> > my logs :-|
>
> It was a rare occurrance, and only really happened if the network card
> never interrupted the kernel for the last transmission for several
> _seconds_, this is rare in practice so I would not be concerned, you'd
> know if you ever hit this because netstat would indicate numerous TCP
> connections stuck in CLOSE, it took a load of around 230 for me to hit
> it over 100baseT on the Sparc.
Interesting, since we have had two CLOSE_WAIT sockets (wasn't it more??)
for around a fortnight now.