[squid-users] measuring latency of squid in different scenarios

[Alex Rousskov]

On 9/28/20 9:19 AM, Rafał Stanilewicz wrote:

> I'd like to get some numbers about squid-introduced latency of getting
> some particular web resource. Is there any benchmarking program I could
> use? I'd like to see what is the current latency of getting the resource
> without any proxying, then of getting the same resource with explicit
> proxy settings, then of implicit (intercepting) proxy option, as well as
> for different options of caching. 

What is the primary goal of collecting those measurements? Let's assume
that the measurements show that Squid adds X% to the median response
time in a particular test scenario. Now what?

* If the primary goal is to just record/report _some_ number and forget
about it, then you can use curl, wget, or ab to generate dumb test
traffic and measure overall response times of primary configurations.
This (mostly pointless from a purely technical point of view) exercise
should not take more than a few hours. It is useful, for example, in
cases where one needs to report some measurements to the management, but
everybody just wants to mark some checkbox on some list.

* If the primary goal is to verify some performance guarantees or tune
Squid performance, then you would need to invest a lot more into these
performance tests. You need stable, reproducible results and
representative traffic pattern(s). I use Web Polygraph
(http://www.web-polygraph.org) for such tests, but it has a steep
learning curve, may need some love to compile in your environment, and
it is a biased recommendation.


HTH,

Alex.