Define slow. With 40 mb cache_mem configured in squid.conf,
you may be paging like crazy. We have 20 mb cache_mem configured
on a 128 meg machine. Squid has stabalized at 106 meg virtual
and 83 meg resident. On our machine, we find we can handle 3-4
page faults/second without impacting performance. When we
were in the 10-20 page faults/second, performance was poor.
We have a machine a little bigger:
AMD K6-300
128 mb
2 Ultra IDE drives, 3 gb of cache on 1, 7 on the other
The box is running freebsd 2.2.8 with the ipnat drivers.
With a peak load of around 17 http requests/second,
we see typical http hit service times of under 40 ms.
Cpu load at this level is less than 50%, and the box
is doing the tranparent redirection as well.
How does that compare with yours?
One thing we found is that we have to flush the ipnat
address maps hourly. Otherwise, we have thousands of
maps and it takes too long to search them, slowing down
performance.
I know someone will bring up using SCSI vs IDE drives. As far
as SCSI vs UIDE goes, I've seen sustained data rates of
8mbytes/second and up to 90 transfers/second from our UIDE
drives. Interrupt utilization seems to be a bit higher than
scsi, as our IDE drives/controllers can only tranfser 16 blocks
at a time. So, I don't think that the IDe drives are your
bottleneck.
On Wed, 3 Feb 1999, Andre L. Silva wrote:
> HI,
>
> I have problems with performance here too.
> I have a SQUID-2.1-patch2 and performance is too slow comparing
> to direct conection (without squid).
> My machine is a P233 MMX with 96MB and 2.5 GB for cache on a IDE
> drive (Quantum Fireball 4.3 GB). In machine only run SendMail, POP, ppp,
> apache (slow trafic, secundary site) and SQUID.
> SQUID has been configured to use 40 MB of memory.
> I have recompiled Squid with option --disable-async-io, and
> nothing hapening.
> Who have any idea ?
> Sincerely,
>
> Andre L. Silva
> andre@starinfo.com.br
Received on Wed Feb 03 1999 - 16:00:48 MST
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:44:20 MST