> I'm pretty sure this is an kernel
> level issue. Increasing the max_tcp_syn_backlog (from 128 to 1024) has made
> it much more difficult to 'stall' apache.
It means that problem is not in apache. And most likely not in the kernel.
You just have average rtt up to client * bandwidth * people who likes
your server too high to be served by queue length 128.
> I'm sure that the wakeone code in 2.3.x has taken care of that issue.
8) How much thousands of conn/sec does your server?
No, your first statements relaibly proves that reason in network.
Increase syn_backlog to a value when the problem disappears
in average. If you hit 1024, you have to add new server.
Another solution is to split load to virtual servers listening
sockets bound to specific addresses.
Alexey
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/