NAME POE::Devel::Benchmarker - Benchmarking POE's performance ( acts more like a smoker ) SYNOPSIS apoc@apoc-x300:~$ cd poe-benchmarker apoc@apoc-x300:~/poe-benchmarker$ perl -MPOE::Devel::Benchmarker -e 'benchmark()' ABSTRACT This package of tools is designed to benchmark POE's performace across different configurations. The current "tests" are: posts This tests how long it takes to post() N times This tests how long it took to dispatch/deliver all the posts enqueued in the post() test This tests how long it took to yield() between 2 states for N times calls This tests how long it took to call() N times alarms This tests how long it took to add N alarms via alarm(), overwriting each other This tests how long it took to add N alarms via alarm_add() and how long it took to delete them all NOTE: alarm_add is not available on all versions of POE! sessions This tests how long it took to create N sessions, and how long it took to destroy them all filehandles This tests how long it took to toggle select_read N times on STDIN and a real filehandle via open() This tests how long it took to toggle select_write N times on STDIN and a real filehandle via open() POE startup time This tests how long it took to start + close N instances of a "virgin" POE without any sessions/etc POE Loops This is actually a "super" test where all of the specific tests is ran against various POE::Loop::XYZ/FOO for comparison POE Assertions This is actually a "super" test where all of the specific tests is ran against POE with/without assertions enabled POE::XS::Queue::Array This is actually a "super" test where all of the specific tests is ran against POE with XS goodness enabled/disabled DESCRIPTION This module is poorly documented now. Please give me some time to properly document it over time :) INSTALLATION Here's a simple outline to get you up to speed quickly. ( and smoking! ) Install CPAN package + dependencies Download+install the POE::Devel::Benchmarker package from CPAN apoc@apoc-x300:~$ cpanp -i POE::Devel::Benchmarker Setup initial directories Go anywhere, and create the "parent" directory where you'll be storing test results + stuff. For this example, I have chosen to use ~/poe-benchmarker: apoc@apoc-x300:~$ mkdir poe-benchmarker apoc@apoc-x300:~$ cd poe-benchmarker apoc@apoc-x300:~/poe-benchmarker$ mkdir poedists results images apoc@apoc-x300:~/poe-benchmarker$ perl -MPOE::Devel::Benchmarker::GetPOEdists -e 'getPOEdists( 1 )' ( go get a coffee while it downloads if you're on a slow link, ha! ) Let 'er rip! At this point you can start running the benchmark! NOTE: the Benchmarker expects everything to be in the "local" directory! apoc@apoc-x300:~$ cd poe-benchmarker apoc@apoc-x300:~/poe-benchmarker$ perl -MPOE::Devel::Benchmarker -e 'benchmark()' ( go sleep or something, this will take a while! ) BENCHMARKING On startup the Benchmarker will look in the "poedists" directory and load all the distributions it sees untarred there. Once that is done it will begin autoprobing for available POE::Loop packages. Once it determines what's available, it will begin the benchmarks. As the Benchmarker goes through the combinations of POE + Eventloop + Assertions + XS::Queue it will dump data into the results directory. The Analyzer module also dumps YAML output in the same place, with the suffix of ".yml" This module exposes only one subroutine, the benchmark() one. You can pass a hashref to it to set various options. Here is a list of the valid options: freshstart => boolean This will tell the Benchmarker to ignore any previous test runs stored in the 'results' directory. benchmark( { freshstart => 1 } ); default: false noxsqueue => boolean This will tell the Benchmarker to force the unavailability of POE::XS::Queue::Array and skip those tests. NOTE: The Benchmarker will set this automatically if it cannot load the module! benchmark( { noxsqueue => 1 } ); default: false noasserts => boolean This will tell the Benchmarker to not run the ASSERT tests. benchmark( { noasserts => 1 } ); default: false litetests => boolean This enables the "lite" tests which will not take up too much time. benchmark( { litetests => 0 } ); default: true quiet => boolean This enables quiet mode which will not print anything to the console except for errors. benchmark( { 'quiet' => 1 } ); default: false loop => csv list or array This overrides the built-in loop detection algorithm which searches for all known loops. There is some "magic" here where you can put a negative sign in front of a loop and we will NOT run that. NOTE: Capitalization is important! benchmark( { 'loop' => 'IO_Poll,Select' } ); # runs only IO::Poll and Select benchmark( { 'loop' => [ qw( Tk Gtk ) ] } ); # runs only Tk and Gtk benchmark( { 'loop' => '-Tk' } ); # runs all available loops EXCEPT for TK Known loops: Event_Lib EV Glib Prima Gtk Kqueue Tk Select IO_Poll poe => csv list or array This overrides the built-in POE version detection algorithm which pulls the POE versions from the 'poedists' directory. There is some "magic" here where you can put a negative sign in front of a version and we will NOT run that. NOTE: The Benchmarker will ignore versions that wasn't found in the directory! benchmark( { 'poe' => '0.35,1.003' } ); # runs on 0.35 and 1.003 benchmark( { 'poe' => [ qw( 0.3009 0.12 ) ] } ); # runs on 0.3009 and 0.12 benchmark( { 'poe' => '-0.35' } ); # runs ALL tests except 0.35 ANALYZING RESULTS Please look at the pretty charts generated by the POE::Devel::Benchmarker::Imager module. EXPORT Automatically exports the benchmark() subroutine. TODO Perl version smoking We should be able to run the benchmark over different Perl versions. This would require some fiddling with our layout + logic. It's not that urgent because the workaround is to simply execute the benchmarker under a different perl binary. It's smart enough to use $^X to be consistent across tests/subprocesses :) Select the EV backend and if you are benchmarking, try it with POE using EV with EV using Glib? :P I'm not sure how to configure the EV "backend" yet too much docs for me to read hah Apocalypse: use EV::Glib; use Glib; use POE; :) Be smarter in smoking timeouts Currently we depend on the litetests option and hardcode some values including the timeout. If your machine is incredibly slow, there's a chance that it could timeout unnecessarily. Please look at the outputs and check to see if there are unusual failures, and inform me. Also, some loops perform badly and take almost forever! /me glares at Gtk... More benchmarks! As usual, me and the crowd in #poe have plenty of ideas for tests. We'll be adding them over time, but if you have an idea please drop me a line and let me know! dngor said there was some benchmarks in the POE svn under trunk/queue... I want a bench that actually tests socket traffic - stream 10MB of traffic over localhost, and time it? LotR and Tapout contributed some samples, let's see if I can integrate them... Add SQLite/DBI/etc support to the Analyzer It would be nice if we could have a local SQLite db to dump our stats into. This would make arbitrary reports much easier than loading raw YAML files and trying to make sense of them, ha! Also, this means somebody can do the smoking and send the SQLite db to another person to generate the graphs, cool! Kqueue loop support As I don't have access to a *BSD box, I cannot really test this. Furthermore, it isn't clear on how I can force/unload this module from POE... Wx loop support I have Wx installed, but it doesn't work. Obviously I don't know how to use Wx ;) If you have experience, please drop me a line on how to do the "right" thing to get Wx loaded under POE. Here's the error: Can't call method "MainLoop" on an undefined value at /usr/local/share/perl/5.8.8/POE/Loop/Wx.pm line 91. XS::Loop support The POE::XS::Loop::* modules theoretically could be tested too. However, they will only work in POE >= 1.003! This renders the concept somewhat moot. Maybe, after POE has progressed some versions we can implement this... SEE ALSO POE AUTHOR Apocalypse BIG THANKS goes to Rocco Caputo for the first benchmarks! COPYRIGHT AND LICENSE Copyright 2008 by Apocalypse This library is free software; you can redistribute it and/or modify it under the same terms as Perl itself.