Re: performance (was Re: cache not up to date)

>
> I've been running benchmarks in our lab using WebStone version 1.1.
> With the distributed load (filelist.ss), which has an average file
> size of 7 K and a maximum file size of 200 K Squid is able to handle
> in the order of 280 connections per second on a 133 MHz P5 running
> BSD/OS or 440 connections per second on a 200 MHz P6. This works out
> to 1.0 million and 1.6 million hits per hour.
>
> With this test the cache hit rate is 100%. I'd expect real life to be
> worse as squid started to have to go to disk.

In fact, it would be orders of magnitude different, at least here in
Germany in academics where a majority of requests cross the extremely
crowded transatlantic links (even to sites just 2 miles away, in fact).

The average transaction time for non-hits is roughly 20 seconds (not
counting error transactions which average at >30 sec), that of cache hit
transfers to reasonably fast machines on the local net is about a second,
 except for modem users whose modem speeds increase the average
transaction time for hits to about 6-10 seconds (average file size going
to modem users is half that of those going to users on our university
LAN;-> ). ICP requests are handled in a few hundreds of milliseconds.

During peak hours, I'm regularly getting Squid to report more than 600
file descriptors used (out of 1024 max!).

It's been my experience that you only need to divide the daily hit rates
by 10 (not 24) to get a good estimate of the peak rate per hour, unless
your server is badly in need of upgrading (i.e. hits peak rate early in
the morning and stays there all day). We get >>20.000 requests per hour
peak rate, about 50% hits, another 10% neighbor hits, about 25%
non-cachable requests (?,/cgi-bin/).
>
> >I'd be curious to hear any other squid performance stats people may
> >have. I'd be particularly interested to hear any comparison between
> >the squid proxy and the Netscape proxy. I haven't evaluated it myself
> >but was hoping someone else has.

Netscape and Squid are both memory-hungry in the setting just described:
Netscape requires much more than 200 processes to handle that load, with
1-2MB per process; Squid requires roughly the same amount of memory (with
system processes about 200MB) to handle the large cache that you need to
drive such a load plus the buffers for the hundreds of simultaneous
connections. In fact, it is my impression that in our environment,
Netscape ultimately required more memory to run effectively since you
absolutely had to avoid any paging activity at all.

With 128MB memory on a Sparc20, we're NOT handling the load properly
running Squid, but we're close.

Andreas

-- 
Andreas Strotmann       / ~~~~~~ \________________A.Strotmann@Uni-Koeln.DE
Universitaet zu Koeln  /| University of Cologne   \
Regionales Rechenzentrum| Regional Computer Center \
Robert-Koch-Str. 10    /|    Tel: +49-221-478-5524 |\   Home: -221-4200663
D-50931  Koeln        __|__  FAX: +49-221-478-5590 |__________~~~~~~~~~~~~