You might want to look at setting the tcp timeout to a smaller
number.. I remember seeing a reference to that somewhere either on this
list or deja, but I can't remember exactly where the value is...
When you do a netstat -n do you see massively more connections that what
are currently persistent from your robots?
Curtis
(yes, It's me Larry)
On Wed, 17 Oct 2001, Larry Creech wrote:
>
> Are there any recommended kernel tweaks for linux 2.4.x series? I've noticed
> that the proxy can handle 1000 robots at 400reps/sec (0.4 req_rate) with the
> robot pconn_use_lmt commented out. If the number of robots is increased an
> the rate is maintained, TCP connections are dropped and the proxy cannot
> keep up. Likewise if persistent connection are re-enabled on the robots, the
> squid/RH7.1 proxy cannot handle even 500 robots at 200reps/sec unless I set
> the pconn_use_lmt to 2. The squid is not overloading the box, but there
> seems to be a maximum number of TCP connections at one time which. To get
> more out of this, what need should be tweaked? Kernel (/proc) and/or squid?
>
> Thanks,
> Larry Creech
>
Received on Wed Oct 17 2001 - 19:29:37 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:02:49 MST