Buffer problems with FreeBSD/squid

I'm having a strange problem with squid-2.3S1 running under FreeBSD 3.3.

We have a Cisco piping traffic to the squid box using WCCP, and in the
past two weeks (after running fine for 6 weeks) the squid box has started
losing network communications. (But can still be accessed from the
console).

The box has MAXUSERS set all the way at 256, 256mb of RAM (not even using
all of it according to top). netstat -m doesn't seem to indicate its
running out of mbufs.

Syslog is reporting:

Jan 12 18:30:45 cache squid[222]: comm_udp_sendto: FD 5, x.x.x.x,
port 53:
 (55) No buffer space available
Jan 12 18:30:45 cache squid[222]: idnsSendQuery: FD 5: sendto: (55) No
buffer space available

Anyone got any ideas? The box already had TCP timeouts reduced from 30 to
3 seconds, the TCP portranges increased, etc.

Thanks.

-marc
Received on Thu Jan 13 2000 - 14:18:41 MST