50% is probably about right for a 550MHz machine at 40 recs/sec.
How many drives did you say you had?  35 threads is probably a bit much 
for one drive.  Actually, the default is great for all but the fastest 
drives--so don't alter the default when configuring.
LFUDA and GDSF are documented by the folks who implemented them:
http://www.hpl.hp.com/techreports/1999/HPL-1999-69.html
http://fog.hpl.external.hp.com/techreports/98/HPL-98-173.html
These links are taken from the squid.conf file.  Note that replacement 
policies are mostly academic--performance differences are very modest, 
as are hit ratio differences.  The benchmarks presented in the papers 
are not realistic enough to really model the world--and I've found 
Polygraph benchmarks shrink the performance + hit rate differences 
between policies to almost nil, assuming a properly configured system 
(i.e. enough RAM for the size of the cache, and enough disk space to 
effectivly cache for the given workload).
khiz code wrote:
> hi joe
> 
> well a guru is a guru and no doubting that
> well yes joe  i did compile squid with the enable time hack option
> ive now compiled a new one without it 
> now my figures are like this
> client_http.requests = 31.505258/sec
> client_http.hits = 14.794440/sec
> cpu_usage = 49.813500%
> i just hv a feeling that even this is abit on the higher side 
> ive compiled 2.4S1 with async-io and ummm 35 threads....
> do u think that this is abit high given the load of abt 40 req/sec ..
> do i need to decrease the no of threads or totally give up async io ..
> 
> also i ve not been able to get sufficient information abt the LFUDA and
> GFUDA replacment plocies and related performance..any pointers to that
> ??
> rgds
> khizcode
> 
> 
> 
> --- Joe Cooper <joe@swelltech.com> wrote:
> 
>>Am I misremembering that the configure options you posted a week or
>>two 
>>back included 'time-hack'?  If you are still running a Squid with 
>>time-hack, recompile without it--it is a hack that does not work as 
>>expected on any platform I know of and leads to severe CPU overusage.
                                   --
                      Joe Cooper <joe@swelltech.com>
                  Affordable Web Caching Proxy Appliances
                         http://www.swelltech.com
Received on Fri Sep 14 2001 - 00:28:19 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:02:10 MST