How to measure hit ratio

Dmitry Kurochkin dmitry.kurochkin at
Wed Jul 13 21:27:43 UTC 2011

Hi Erico.

On Wed, 13 Jul 2011 17:52:40 -0300, Erico Augusto Cavalcanti Guedes <eacg at> wrote:
> Dears,
> is there a way to measure the hit ratio of a cache experiment using
> polygraph tools, resources...


> I'm loading a cache cluster of 4 nodes, and find a way to process squid
> access.log files together to establish  isn't a trivial task.
> Lab is composed of 6 machines: One to execute polygraph-server, another to
> polygraph-client and 4 on cache cluster. If there is a way to gather hit
> ratio on polygraph-client machine, I'll spend less time and, who knows,
> watch Brazilian soccer team tonight! It's missing 4 hours to match start!
> I'm looking for the solution in Google, Search PG docs and with you...

You should start with reading [1] and [2].  That should give you
information on how to run basic tests.  Hit ratio is reported on client
console during the test, see [3] for console output format.  A more
detailed information can be obtained from Polygraph logs (--log option
for polygraph-client(1) and polygraph-server(1)).  The simplest way to
do that is using polygraph-reporter(1) tool, documented at [4].  Just
run the reporter on both client and server logs, open index.html page in
the generated report, and click "Hit ratios" link.

While 4 hours should be enough to run some basic tests, I am afraid it
would take much more time to create and run a complex workload.

Note that to test multiple proxies you need to set them through PGL
Robot.*proxies fields, --proxy command line option accepts only a single
proxy address.  Moreover proxy selection is sticky, which means that
each Robot selects a proxy on startup and uses it during the entire
test.  That means you need to use multiple Robots to test several
proxies.  See [5] for more information on PGL Robot.*proxies fields.

I hope this would help and keep you busy till the soccer match :)



> Best regards,
> Erico.
> _______________________________________________
> Users mailing list
> Users at

More information about the Users mailing list