Hi,
I had a similar problem caused by MS clients and MS proxy connecting over a
slow link. The machine was getting the SYN packets, but the TCP handshake was
not completing (an unintentional SYN flood, dur to high packet loss on a WAN
link). The TCP stack has a limit to the number of incoming connect that
requests that are in this state. The system would also recover by itself after
a while.
I found that even though squid was not accepting connects, I could still telnet
to the box, and the CPU load was very low, but a netstat -a showed about 127
connections to squid in a SYN_ACK state.
The fix was to increase the length of the tcp_syn_backlog by doing echo 256 >
/proc/sys/net/ipv4/tcp_max_syn_backlog.
256 seems enough for us now (128 is the default).
I don't know if this is your problem, as your users are on a LAN, but seeing
that you can still get to the proxy on the second interface, I suspect it is
the same problem (there is one queue for each socket, and I guess the 2
interfaces mean 2 queues?).
Hope this helps.
Sarah Rodriguez wrote:
> Hi there!
>
> My squid server seems not to respond to requests after sometime of excellent
> operation.
>
> I have the latest linux server (slackware 4.0) running squids latest
> (version 2.2.STABLE4 ). There are about 100 pcs using this with an IE 4.0
> browser. The server runs on an Intel Pentium Pro 200 with cache size of
> about 3 GB. It has two NICs .. one is connected to the router (202.8.254.3)
> while the other to our LAN (172.16.12.254) where the PCs are connected. My
> problem is after sometime, the squid server seems not to respond anymore (as
> if it hanged) ... and it polls ... and after sometime .. a message pops up
> saying "A connection with the server could not be established". This
> normally happens if the proxy server is down.
>
-- Darren Steven Applications Specialist Networking Tasmania Telstra Australia Ph.1800 813 302Received on Wed Sep 15 1999 - 19:21:26 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:48:25 MST