W dniu 24.09.2016 o 00:15, Jakub Narębski pisze:
> W dniu 23.09.2016 o 11:08, Ian Kelling napisał:
>> After curling blob view of unhighlighted large and small text
>> files of perl code and license text 100 times each on a local
>> Apache/2.4.23 (Debian) instance, it's logs indicate +-1% difference in
>> request time for all file types.
> Also, "curling" is not the word I would like to see. I would say:
> Simple benchmark comparing performance of 'blob' view of files without
> syntax highlighting in gitweb before and after this change indicates
> ±1% difference in request time for all file types. Benchmark was
> performed on local instance on Debian, using Apache/2.4.23 web server
> and CGI/PSGI/FCGI/mod_perl.
> ^^^^^^^^^^^^^^^^^^^^^^--- select one
> Or something like that; I'm not sure how detailed this should be.
> But it is nice to have such benchmark in the commit message.
Sidenote: this way of benchmarking of gitweb falls between two ways of
doing a benchmark.
The first method is to simply run gitweb as a standalone script, passing
its parameters in CGI environment variables; just like the test suite
does it. You would 'time' / 'times' it a few times, drop outliers, and
take average or a median. With this method you don't even need to set
up a web server.
The second is to use a specialized program to benchmark the server-side
of a web page, for example 'ab' (ApacheBench), httperf, curl-loader
or JMeter. The first one is usually distributed together with Apache
web server, so you probably have it installed already. Those tools
provide timing statistics.
> Note that the performance loss might be quite higher on MS Windows, with
> its higher cost of fork. But then they probably do not configure
> server-side highligher anyway.