Squid 2.1.PATCH2 on FreeBSD 3.1-STABLE and sockets in CLOSING state.

We've been running Squid in http accelerator mode. We've been serving
requests for about a month, now, at about 300ppm, about 2/3 of those
being cache hits.

Last night, we had a large download (1.5M) which unexplainedly went very
slow (at the end it was going around 8KB/s, across a T1 connection with
two hops between myself and the Squid box). Letting the download proceed
to completion fixed the problem, and now it goes as fast as can be
expected (>100KB/s).

The only suspicious item I was able to find is that netstat shows an
excessive number of Squid sockets in the CLOSING state, around 1100 as of
now. I scanned the logfiles for a couple of the remote addresses, and
found them at varying distances in the past - though I didn't find a
number of them, and our logfiles go back about ten days. So those sockets
seem to be accumulating.

Any ideas?

-- 
scott hess <scott@doubleu.com> (408) 739-8858 http://www.doubleu.com/
<Favorite unused computer book title: The Compleat Demystified Idiots
  Guide to the Zen of Dummies in a Nutshell in Seven Days, Unleashed>