Call to Benchmark PSGI!

[Dylan Hardison]

Given multiple runs, it seems that Gazelle is only a little faster
then a correctly-configured mod_perl. Nothing to make headlines over.
PSGI still brings plenty of advantages, like not requiring apache and
more pleasant dev environment.

apache mod_perl, size limit = 750_000, 10 workers:

Concurrency Level:      8
Time taken for tests:   110.009 seconds
Complete requests:      4000
Failed requests:        0
Total transferred:      114804000 bytes
HTML transferred:       113272000 bytes
Requests per second:    36.36 [#/sec] (mean)
Time per request:       220.018 [ms] (mean)
Time per request:       27.502 [ms] (mean, across all concurrent requests)
Transfer rate:          1019.13 [Kbytes/sec] received

Gazelle, max-workers: 10:

Concurrency Level:      8
Time taken for tests:   95.263 seconds
Complete requests:      4000
Failed requests:        0
Total transferred:      114440000 bytes
HTML transferred:       113140000 bytes
Requests per second:    41.99 [#/sec] (mean)
Time per request:       190.526 [ms] (mean)
Time per request:       23.816 [ms] (mean, across all concurrent requests)
Transfer rate:          1173.15 [Kbytes/sec] received

On Wed, Jan 20, 2016 at 8:33 AM, Denis Roy <denis.roy at eclipse.org> wrote:
> We've been using 300_000 as the magic number for years now, and I find
> bz performance to be quite acceptable. We do have three web heads though.
>
> I can't wait to see your benchmark against psgi.
>
> Denis
>
>
>
> On 01/20/2016 12:17 AM, Dylan Hardison wrote:
>> I knew these numbers were too good to be true -- it turns out the
>> stock mod_perl config has such a low Apache::SizeLimit that the
>> process gets killed after a few requests. I've re-opened
>> https://bugzilla.mozilla.org/show_bug.cgi?id=731589 and the new
>> default will be a more more reasonable 250_000.
>>
>> I'm still keen on benchmarks, and I'll post my revised ones shortly. I
>> recommend editing mod_perl.pl and setting the size limit to 250_000 or
>> 750_000,
>> or wait for Bug 731589 to land.
>>
>> On Tue, Jan 19, 2016 at 12:33 PM, Dylan Hardison <dylan at mozilla.com> wrote:
>>> I think PSGI is *ridiculously* faster than mod_perl
>>>
>>>
>>> Given a simple benchmark of calling apache bench against show_bug?id=1
>>> (with a simple testing db, mind you) I get 12 requests a second under PSGI
>>> (using HTTP::Server::PSGI) and 4 requests/second with mod_perl
>>>
>>> So mod_perl is about twice as fast CGI, but PSGI is three times faster
>>> than mod_perl.
>>>
>>> https://gist.github.com/dylanwh/d5c5498188c3866b7a03d
>>>
>>> So a call to people out there that have a working mod_perl setup ---
>>> please benchmark
>>> master on both mod_perl and PSGI and post the results here!
>>>
>>> Regards,
>>>
>>> Dylan.
>> -
>> To view or change your list settings, click here:
>> <http://bugzilla.org/cgi-bin/mj_wwwusr?user=denis.roy@eclipse.org>
>>
> -
> To view or change your list settings, click here:
> <http://bugzilla.org/cgi-bin/mj_wwwusr?user=dylan@mozilla.com>