This is what i have done to tune the non-caching squid box. Did i miss out
anything?
/etc/sysctl.conf
net.ipv4.tcp_keepalive_probes = 8
net.ipv4.tcp_syn_retries = 2
net.ipv4.tcp_synack_retries = 2
net.ipv4.tcp_keepalive_time = 600
net.ipv4.tcp_wmem = 8192 16384 131072
net.ipv4.tcp_rmem = 65536 65536 174760
net.ipv4.tcp_fin_timeout = 4
When the load is about 100 req/s, i got the following stats, is this
consider normal?
fd 1200
total connection 14000
est connection 4000
Really dream of hitting 700req/s... :)
On Wed, 8 Jan 2003, Henrik Nordstrom wrote:
> Wei Keong wrote:
> >
> > reducing cache_mem (256M -> 128M) and fd (8152 - 4096) doesn't improve the
> > proformance much. my squid becomes very slow (ave response time = 1s, cpu
> > = 100%) when req/s = 100.
> >
> > tried cutting down url_regex also does help much... is 100 req/s the 'max'
> > load for PIII 1GHz, 2GB RAM system?
>
> No, but you may need to retune your TCP kernel to allow such high rate
> of outgoing TCP connections.
>
> Also, if you are doing caching then proper tuning of your disk system is
> needed.
>
> A non-caching Squid on PIII 1GHz can handle ca 700 req/s.
>
> Regards
> Henrik
>
Received on Wed Jan 08 2003 - 21:11:13 MST
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:12:35 MST