Re: [SQU] speed limits

I can refute that. Different versions of Squid max out at different
places, and the hardware it runs on also makes a big difference.

Async i/o Squid is faster than DiskD Squid which is faster than standard
single process Squid. Squid on ReiserFS is faster than Squid on
SoftUpdates which is faster than Squid on a traditional UFS (or ext2)
mounted noatime and nobody with performance in mind will be running UFS
without mounting it noatime.

We have dual disk SCSI 10k boxes that max out at about 160-180 on
Polygraph benchmarks (it depends on which workload). They use ReiserFS,
async i/o Squid 2.2.STABLE5+hno with 32 threads, and have 900MHz Athlon
processor, 512MB RAM, and Ultra 160 SCSI controllers. Squid 2.4 running
async looks like it's going to up the performance by a small margin on
these boxes (and Squid+reiser_raw that will probably work it's way into
2.5 increases it by about 20-30%).

Depending on your processor and compile time options in Squid, 85 should
not be overly hard for a 6 disk box...assuming you have the RAM to
support that much cache_dir and that many simultaneous clients.

Jon Mansey wrote:

> Hi,
>
> I have heard that squid has a "theoretical" access rate brickwall at
> around 100 per second. Does this include ICP and HTTP accesses or just
> HTTP?
>
> Can anyone refute this number? I have a 6 spindle box doing around
> 85reqs/sec total ICP+HTTP and Im wondering if it isnt about maxed.

                                   --
                      Joe Cooper <joe@swelltech.com>
                  Affordable Web Caching Proxy Appliances
                         http://www.swelltech.com

--
To unsubscribe, see http://www.squid-cache.org/mailing-lists.html