[ViewVC] Diff of: cvs/AnyEvent/lib/AnyEvent.pm

Comparing AnyEvent/lib/AnyEvent.pm (file contents):
Revision 1.90 by root, Fri Apr 25 14:24:29 2008 UTC vs.
Revision 1.105 by root, Thu May 1 12:35:54 2008 UTC

…		…
65	technically possible.	65	technically possible.
66		66
67	Of course, if you want lots of policy (this can arguably be somewhat	67	Of course, if you want lots of policy (this can arguably be somewhat
68	useful) and you want to force your users to use the one and only event	68	useful) and you want to force your users to use the one and only event
69	model, you should I<not> use this module.	69	model, you should I<not> use this module.
70
71		70
72	=head1 DESCRIPTION	71	=head1 DESCRIPTION
73		72
74	L<AnyEvent> provides an identical interface to multiple event loops. This	73	L<AnyEvent> provides an identical interface to multiple event loops. This
75	allows module authors to utilise an event loop without forcing module	74	allows module authors to utilise an event loop without forcing module
…		…
298	# do something else, then wait for process exit	297	# do something else, then wait for process exit
299	$done->wait;	298	$done->wait;
300		299
301	=head2 CONDITION VARIABLES	300	=head2 CONDITION VARIABLES
302		301
		302	If you are familiar with some event loops you will know that all of them
		303	require you to run some blocking "loop", "run" or similar function that
		304	will actively watch for new events and call your callbacks.
		305
		306	AnyEvent is different, it expects somebody else to run the event loop and
		307	will only block when necessary (usually when told by the user).
		308
		309	The instrument to do that is called a "condition variable", so called
		310	because they represent a condition that must become true.
		311
303	Condition variables can be created by calling the C<< AnyEvent->condvar >>	312	Condition variables can be created by calling the C<< AnyEvent->condvar
304	method without any arguments.	313	>> method, usually without arguments. The only argument pair allowed is
		314	C<cb>, which specifies a callback to be called when the condition variable
		315	becomes true.
305		316
306	A condition variable waits for a condition - precisely that the C<<	317	After creation, the conditon variable is "false" until it becomes "true"
307	->broadcast >> method has been called.	318	by calling the C<broadcast> method.
308		319
309	They are very useful to signal that a condition has been fulfilled, for	320	Condition variables are similar to callbacks, except that you can
		321	optionally wait for them. They can also be called merge points - points
		322	in time where multiple outstandign events have been processed. And yet
		323	another way to call them is transations - each condition variable can be
		324	used to represent a transaction, which finishes at some point and delivers
		325	a result.
		326
		327	Condition variables are very useful to signal that something has finished,
310	example, if you write a module that does asynchronous http requests,	328	for example, if you write a module that does asynchronous http requests,
311	then a condition variable would be the ideal candidate to signal the	329	then a condition variable would be the ideal candidate to signal the
312	availability of results.	330	availability of results. The user can either act when the callback is
		331	called or can synchronously C<< ->wait >> for the results.
313		332
314	You can also use condition variables to block your main program until	333	You can also use them to simulate traditional event loops - for example,
315	an event occurs - for example, you could C<< ->wait >> in your main	334	you can block your main program until an event occurs - for example, you
316	program until the user clicks the Quit button in your app, which would C<<	335	could C<< ->wait >> in your main program until the user clicks the Quit
317	->broadcast >> the "quit" event.	336	button of your app, which would C<< ->broadcast >> the "quit" event.
318		337
319	Note that condition variables recurse into the event loop - if you have	338	Note that condition variables recurse into the event loop - if you have
320	two pirces of code that call C<< ->wait >> in a round-robbin fashion, you	339	two pieces of code that call C<< ->wait >> in a round-robbin fashion, you
321	lose. Therefore, condition variables are good to export to your caller, but	340	lose. Therefore, condition variables are good to export to your caller, but
322	you should avoid making a blocking wait yourself, at least in callbacks,	341	you should avoid making a blocking wait yourself, at least in callbacks,
323	as this asks for trouble.	342	as this asks for trouble.
324		343
325	This object has two methods:	344	Condition variables are represented by hash refs in perl, and the keys
		345	used by AnyEvent itself are all named C<_ae_XXX> to make subclassing
		346	easy (it is often useful to build your own transaction class on top of
		347	AnyEvent). To subclass, use C<AnyEvent::CondVar> as base class and call
		348	it's C<new> method in your own C<new> method.
326		349
327	=over 4	350	There are two "sides" to a condition variable - the "producer side" which
328		351	eventually calls C<< -> broadcast >>, and the "consumer side", which waits
329	=item $cv->wait	352	for the broadcast to occur.
330
331	Wait (blocking if necessary) until the C<< ->broadcast >> method has been
332	called on c<$cv>, while servicing other watchers normally.
333
334	You can only wait once on a condition - additional calls will return
335	immediately.
336
337	Not all event models support a blocking wait - some die in that case
338	(programs might want to do that to stay interactive), so I<if you are
339	using this from a module, never require a blocking wait>, but let the
340	caller decide whether the call will block or not (for example, by coupling
341	condition variables with some kind of request results and supporting
342	callbacks so the caller knows that getting the result will not block,
343	while still suppporting blocking waits if the caller so desires).
344
345	Another reason I<never> to C<< ->wait >> in a module is that you cannot
346	sensibly have two C<< ->wait >>'s in parallel, as that would require
347	multiple interpreters or coroutines/threads, none of which C<AnyEvent>
348	can supply (the coroutine-aware backends L<AnyEvent::Impl::CoroEV> and
349	L<AnyEvent::Impl::CoroEvent> explicitly support concurrent C<< ->wait >>'s
350	from different coroutines, however).
351
352	=item $cv->broadcast
353
354	Flag the condition as ready - a running C<< ->wait >> and all further
355	calls to C<wait> will (eventually) return after this method has been
356	called. If nobody is waiting the broadcast will be remembered..
357
358	=back
359		353
360	Example:	354	Example:
361		355
362	# wait till the result is ready	356	# wait till the result is ready
363	my $result_ready = AnyEvent->condvar;	357	my $result_ready = AnyEvent->condvar;
…		…
369	my $w = AnyEvent->timer (	363	my $w = AnyEvent->timer (
370	after => 1,	364	after => 1,
371	cb => sub { $result_ready->broadcast },	365	cb => sub { $result_ready->broadcast },
372	);	366	);
373		367
374	# this "blocks" (while handling events) till the watcher	368	# this "blocks" (while handling events) till the callback
375	# calls broadcast	369	# calls broadcast
376	$result_ready->wait;	370	$result_ready->wait;
		371
		372	=head3 METHODS FOR PRODUCERS
		373
		374	These methods should only be used by the producing side, i.e. the
		375	code/module that eventually broadcasts the signal. Note that it is also
		376	the producer side which creates the condvar in most cases, but it isn't
		377	uncommon for the consumer to create it as well.
		378
		379	=over 4
		380
		381	=item $cv->broadcast (...)
		382
		383	Flag the condition as ready - a running C<< ->wait >> and all further
		384	calls to C<wait> will (eventually) return after this method has been
		385	called. If nobody is waiting the broadcast will be remembered.
		386
		387	If a callback has been set on the condition variable, it is called
		388	immediately from within broadcast.
		389
		390	Any arguments passed to the C<broadcast> call will be returned by all
		391	future C<< ->wait >> calls.
		392
		393	=item $cv->croak ($error)
		394
		395	Similar to broadcast, but causes all call's wait C<< ->wait >> to invoke
		396	C<Carp::croak> with the given error message/object/scalar.
		397
		398	This can be used to signal any errors to the condition variable
		399	user/consumer.
		400
		401	=item $cv->begin ([group callback])
		402
		403	=item $cv->end
		404
		405	These two methods can be used to combine many transactions/events into
		406	one. For example, a function that pings many hosts in parallel might want
		407	to use a condition variable for the whole process.
		408
		409	Every call to C<< ->begin >> will increment a counter, and every call to
		410	C<< ->end >> will decrement it. If the counter reaches C<0> in C<< ->end
		411	>>, the (last) callback passed to C<begin> will be executed. That callback
		412	is I<supposed> to call C<< ->broadcast >>, but that is not required. If no
		413	callback was set, C<broadcast> will be called without any arguments.
		414
		415	Let's clarify this with the ping example:
		416
		417	my $cv = AnyEvent->condvar;
		418
		419	my %result;
		420	$cv->begin (sub { $cv->broadcast (\%result) });
		421
		422	for my $host (@list_of_hosts) {
		423	$cv->begin;
		424	ping_host_then_call_callback $host, sub {
		425	$result{$host} = ...;
		426	$cv->end;
		427	};
		428	}
		429
		430	$cv->end;
		431
		432	This code fragment supposedly pings a number of hosts and calls
		433	C<broadcast> after results for all then have have been gathered - in any
		434	order. To achieve this, the code issues a call to C<begin> when it starts
		435	each ping request and calls C<end> when it has received some result for
		436	it. Since C<begin> and C<end> only maintain a counter, the order in which
		437	results arrive is not relevant.
		438
		439	There is an additional bracketing call to C<begin> and C<end> outside the
		440	loop, which serves two important purposes: first, it sets the callback
		441	to be called once the counter reaches C<0>, and second, it ensures that
		442	broadcast is called even when C<no> hosts are being pinged (the loop
		443	doesn't execute once).
		444
		445	This is the general pattern when you "fan out" into multiple subrequests:
		446	use an outer C<begin>/C<end> pair to set the callback and ensure C<end>
		447	is called at least once, and then, for each subrequest you start, call
		448	C<begin> and for eahc subrequest you finish, call C<end>.
		449
		450	=back
		451
		452	=head3 METHODS FOR CONSUMERS
		453
		454	These methods should only be used by the consuming side, i.e. the
		455	code awaits the condition.
		456
		457	=item $cv->wait
		458
		459	Wait (blocking if necessary) until the C<< ->broadcast >> or C<< ->croak
		460	>> methods have been called on c<$cv>, while servicing other watchers
		461	normally.
		462
		463	You can only wait once on a condition - additional calls are valid but
		464	will return immediately.
		465
		466	If an error condition has been set by calling C<< ->croak >>, then this
		467	function will call C<croak>.
		468
		469	In list context, all parameters passed to C<broadcast> will be returned,
		470	in scalar context only the first one will be returned.
		471
		472	Not all event models support a blocking wait - some die in that case
		473	(programs might want to do that to stay interactive), so I<if you are
		474	using this from a module, never require a blocking wait>, but let the
		475	caller decide whether the call will block or not (for example, by coupling
		476	condition variables with some kind of request results and supporting
		477	callbacks so the caller knows that getting the result will not block,
		478	while still suppporting blocking waits if the caller so desires).
		479
		480	Another reason I<never> to C<< ->wait >> in a module is that you cannot
		481	sensibly have two C<< ->wait >>'s in parallel, as that would require
		482	multiple interpreters or coroutines/threads, none of which C<AnyEvent>
		483	can supply (the coroutine-aware backends L<AnyEvent::Impl::CoroEV> and
		484	L<AnyEvent::Impl::CoroEvent> explicitly support concurrent C<< ->wait >>'s
		485	from different coroutines, however).
		486
		487	You can ensure that C<< -wait >> never blocks by setting a callback and
		488	only calling C<< ->wait >> from within that callback (or at a later
		489	time). This will work even when the event loop does not support blocking
		490	waits otherwise.
		491
		492	=back
377		493
378	=head1 GLOBAL VARIABLES AND FUNCTIONS	494	=head1 GLOBAL VARIABLES AND FUNCTIONS
379		495
380	=over 4	496	=over 4
381		497
…		…
391		507
392	AnyEvent::Impl::CoroEV based on Coro::EV, best choice.	508	AnyEvent::Impl::CoroEV based on Coro::EV, best choice.
393	AnyEvent::Impl::CoroEvent based on Coro::Event, second best choice.	509	AnyEvent::Impl::CoroEvent based on Coro::Event, second best choice.
394	AnyEvent::Impl::EV based on EV (an interface to libev, best choice).	510	AnyEvent::Impl::EV based on EV (an interface to libev, best choice).
395	AnyEvent::Impl::Event based on Event, second best choice.	511	AnyEvent::Impl::Event based on Event, second best choice.
		512	AnyEvent::Impl::Perl pure-perl implementation, fast and portable.
396	AnyEvent::Impl::Glib based on Glib, third-best choice.	513	AnyEvent::Impl::Glib based on Glib, third-best choice.
397	AnyEvent::Impl::Perl pure-perl implementation, inefficient but portable.
398	AnyEvent::Impl::Tk based on Tk, very bad choice.	514	AnyEvent::Impl::Tk based on Tk, very bad choice.
399	AnyEvent::Impl::Qt based on Qt, cannot be autoprobed (see its docs).	515	AnyEvent::Impl::Qt based on Qt, cannot be autoprobed (see its docs).
400	AnyEvent::Impl::EventLib based on Event::Lib, leaks memory and worse.	516	AnyEvent::Impl::EventLib based on Event::Lib, leaks memory and worse.
401	AnyEvent::Impl::POE based on POE, not generic enough for full support.	517	AnyEvent::Impl::POE based on POE, not generic enough for full support.
402		518
…		…
458		574
459	You can chose to use a rather inefficient pure-perl implementation by	575	You can chose to use a rather inefficient pure-perl implementation by
460	loading the C<AnyEvent::Impl::Perl> module, which gives you similar	576	loading the C<AnyEvent::Impl::Perl> module, which gives you similar
461	behaviour everywhere, but letting AnyEvent chose is generally better.	577	behaviour everywhere, but letting AnyEvent chose is generally better.
462		578
		579	=head1 OTHER MODULES
		580
		581	The following is a non-exhaustive list of additional modules that use
		582	AnyEvent and can therefore be mixed easily with other AnyEvent modules
		583	in the same program. Some of the modules come with AnyEvent, some are
		584	available via CPAN.
		585
		586	=over 4
		587
		588	=item L<AnyEvent::Util>
		589
		590	Contains various utility functions that replace often-used but blocking
		591	functions such as C<inet_aton> by event-/callback-based versions.
		592
		593	=item L<AnyEvent::Handle>
		594
		595	Provide read and write buffers and manages watchers for reads and writes.
		596
		597	=item L<AnyEvent::Socket>
		598
		599	Provides a means to do non-blocking connects, accepts etc.
		600
		601	=item L<AnyEvent::HTTPD>
		602
		603	Provides a simple web application server framework.
		604
		605	=item L<AnyEvent::DNS>
		606
		607	Provides asynchronous DNS resolver capabilities, beyond what
		608	L<AnyEvent::Util> offers.
		609
		610	=item L<AnyEvent::FastPing>
		611
		612	The fastest ping in the west.
		613
		614	=item L<Net::IRC3>
		615
		616	AnyEvent based IRC client module family.
		617
		618	=item L<Net::XMPP2>
		619
		620	AnyEvent based XMPP (Jabber protocol) module family.
		621
		622	=item L<Net::FCP>
		623
		624	AnyEvent-based implementation of the Freenet Client Protocol, birthplace
		625	of AnyEvent.
		626
		627	=item L<Event::ExecFlow>
		628
		629	High level API for event-based execution flow control.
		630
		631	=item L<Coro>
		632
		633	Has special support for AnyEvent.
		634
		635	=item L<IO::Lambda>
		636
		637	The lambda approach to I/O - don't ask, look there. Can use AnyEvent.
		638
		639	=item L<IO::AIO>
		640
		641	Truly asynchronous I/O, should be in the toolbox of every event
		642	programmer. Can be trivially made to use AnyEvent.
		643
		644	=item L<BDB>
		645
		646	Truly asynchronous Berkeley DB access. Can be trivially made to use
		647	AnyEvent.
		648
		649	=back
		650
463	=cut	651	=cut
464		652
465	package AnyEvent;	653	package AnyEvent;
466		654
467	no warnings;	655	no warnings;
…		…
482	my @models = (	670	my @models = (
483	[Coro::EV:: => AnyEvent::Impl::CoroEV::],	671	[Coro::EV:: => AnyEvent::Impl::CoroEV::],
484	[Coro::Event:: => AnyEvent::Impl::CoroEvent::],	672	[Coro::Event:: => AnyEvent::Impl::CoroEvent::],
485	[EV:: => AnyEvent::Impl::EV::],	673	[EV:: => AnyEvent::Impl::EV::],
486	[Event:: => AnyEvent::Impl::Event::],	674	[Event:: => AnyEvent::Impl::Event::],
487	[Glib:: => AnyEvent::Impl::Glib::],
488	[Tk:: => AnyEvent::Impl::Tk::],	675	[Tk:: => AnyEvent::Impl::Tk::],
489	[Wx:: => AnyEvent::Impl::POE::],	676	[Wx:: => AnyEvent::Impl::POE::],
490	[Prima:: => AnyEvent::Impl::POE::],	677	[Prima:: => AnyEvent::Impl::POE::],
491	[AnyEvent::Impl::Perl:: => AnyEvent::Impl::Perl::],	678	[AnyEvent::Impl::Perl:: => AnyEvent::Impl::Perl::],
492	# everything below here will not be autoprobed as the pureperl backend should work everywhere	679	# everything below here will not be autoprobed as the pureperl backend should work everywhere
		680	[Glib:: => AnyEvent::Impl::Glib::],
493	[Event::Lib:: => AnyEvent::Impl::EventLib::], # too buggy	681	[Event::Lib:: => AnyEvent::Impl::EventLib::], # too buggy
494	[Qt:: => AnyEvent::Impl::Qt::], # requires special main program	682	[Qt:: => AnyEvent::Impl::Qt::], # requires special main program
495	[POE::Kernel:: => AnyEvent::Impl::POE::], # lasciate ogni speranza	683	[POE::Kernel:: => AnyEvent::Impl::POE::], # lasciate ogni speranza
496	);	684	);
497		685
…		…
894	});	1082	});
895		1083
896	$quit->wait;	1084	$quit->wait;
897		1085
898		1086
899	=head1 BENCHMARK	1087	=head1 BENCHMARKS
900		1088
901	To give you an idea of the performance and overheads that AnyEvent adds	1089	To give you an idea of the performance and overheads that AnyEvent adds
902	over the event loops themselves (and to give you an impression of the	1090	over the event loops themselves and to give you an impression of the speed
903	speed of various event loops), here is a benchmark of various supported	1091	of various event loops I prepared some benchmarks.
904	event models natively and with anyevent. The benchmark creates a lot of	1092
905	timers (with a zero timeout) and I/O watchers (watching STDOUT, a pty, to	1093	=head2 BENCHMARKING ANYEVENT OVERHEAD
		1094
		1095	Here is a benchmark of various supported event models used natively and
		1096	through anyevent. The benchmark creates a lot of timers (with a zero
		1097	timeout) and I/O watchers (watching STDOUT, a pty, to become writable,
906	become writable, which it is), lets them fire exactly once and destroys	1098	which it is), lets them fire exactly once and destroys them again.
907	them again.
908		1099
909	Rewriting the benchmark to use many different sockets instead of using	1100	Source code for this benchmark is found as F<eg/bench> in the AnyEvent
910	the same filehandle for all I/O watchers results in a much longer runtime	1101	distribution.
911	(socket creation is expensive), but qualitatively the same figures, so it
912	was not used.
913		1102
914	=head2 Explanation of the columns	1103	=head3 Explanation of the columns
915		1104
916	I<watcher> is the number of event watchers created/destroyed. Since	1105	I<watcher> is the number of event watchers created/destroyed. Since
917	different event models feature vastly different performances, each event	1106	different event models feature vastly different performances, each event
918	loop was given a number of watchers so that overall runtime is acceptable	1107	loop was given a number of watchers so that overall runtime is acceptable
919	and similar between tested event loop (and keep them from crashing): Glib	1108	and similar between tested event loop (and keep them from crashing): Glib
…		…
935	signal the end of this phase.	1124	signal the end of this phase.
936		1125
937	I<destroy> is the time, in microseconds, that it takes to destroy a single	1126	I<destroy> is the time, in microseconds, that it takes to destroy a single
938	watcher.	1127	watcher.
939		1128
940	=head2 Results	1129	=head3 Results
941		1130
942	name watchers bytes create invoke destroy comment	1131	name watchers bytes create invoke destroy comment
943	EV/EV 400000 244 0.56 0.46 0.31 EV native interface	1132	EV/EV 400000 244 0.56 0.46 0.31 EV native interface
944	EV/Any 100000 244 2.50 0.46 0.29 EV + AnyEvent watchers	1133	EV/Any 100000 244 2.50 0.46 0.29 EV + AnyEvent watchers
945	CoroEV/Any 100000 244 2.49 0.44 0.29 coroutines + Coro::Signal	1134	CoroEV/Any 100000 244 2.49 0.44 0.29 coroutines + Coro::Signal
946	Perl/Any 100000 513 4.92 0.87 1.12 pure perl implementation	1135	Perl/Any 100000 513 4.92 0.87 1.12 pure perl implementation
947	Event/Event 16000 516 31.88 31.30 0.85 Event native interface	1136	Event/Event 16000 516 31.88 31.30 0.85 Event native interface
948	Event/Any 16000 936 39.17 33.63 1.43 Event + AnyEvent watchers	1137	Event/Any 16000 590 35.75 31.42 1.08 Event + AnyEvent watchers
949	Glib/Any 16000 1357 98.22 12.41 54.00 quadratic behaviour	1138	Glib/Any 16000 1357 98.22 12.41 54.00 quadratic behaviour
950	Tk/Any 2000 1860 26.97 67.98 14.00 SEGV with >> 2000 watchers	1139	Tk/Any 2000 1860 26.97 67.98 14.00 SEGV with >> 2000 watchers
951	POE/Event 2000 6644 108.64 736.02 14.73 via POE::Loop::Event	1140	POE/Event 2000 6644 108.64 736.02 14.73 via POE::Loop::Event
952	POE/Select 2000 6343 94.13 809.12 565.96 via POE::Loop::Select	1141	POE/Select 2000 6343 94.13 809.12 565.96 via POE::Loop::Select
953		1142
954	=head2 Discussion	1143	=head3 Discussion
955		1144
956	The benchmark does I<not> measure scalability of the event loop very	1145	The benchmark does I<not> measure scalability of the event loop very
957	well. For example, a select-based event loop (such as the pure perl one)	1146	well. For example, a select-based event loop (such as the pure perl one)
958	can never compete with an event loop that uses epoll when the number of	1147	can never compete with an event loop that uses epoll when the number of
959	file descriptors grows high. In this benchmark, all events become ready at	1148	file descriptors grows high. In this benchmark, all events become ready at
960	the same time, so select/poll-based implementations get an unnatural speed	1149	the same time, so select/poll-based implementations get an unnatural speed
961	boost.	1150	boost.
		1151
		1152	Also, note that the number of watchers usually has a nonlinear effect on
		1153	overall speed, that is, creating twice as many watchers doesn't take twice
		1154	the time - usually it takes longer. This puts event loops tested with a
		1155	higher number of watchers at a disadvantage.
		1156
		1157	To put the range of results into perspective, consider that on the
		1158	benchmark machine, handling an event takes roughly 1600 CPU cycles with
		1159	EV, 3100 CPU cycles with AnyEvent's pure perl loop and almost 3000000 CPU
		1160	cycles with POE.
962		1161
963	C<EV> is the sole leader regarding speed and memory use, which are both	1162	C<EV> is the sole leader regarding speed and memory use, which are both
964	maximal/minimal, respectively. Even when going through AnyEvent, it uses	1163	maximal/minimal, respectively. Even when going through AnyEvent, it uses
965	far less memory than any other event loop and is still faster than Event	1164	far less memory than any other event loop and is still faster than Event
966	natively.	1165	natively.
…		…
989	file descriptor is dup()ed for each watcher. This shows that the dup()	1188	file descriptor is dup()ed for each watcher. This shows that the dup()
990	employed by some adaptors is not a big performance issue (it does incur a	1189	employed by some adaptors is not a big performance issue (it does incur a
991	hidden memory cost inside the kernel which is not reflected in the figures	1190	hidden memory cost inside the kernel which is not reflected in the figures
992	above).	1191	above).
993		1192
994	C<POE>, regardless of underlying event loop (whether using its pure	1193	C<POE>, regardless of underlying event loop (whether using its pure perl
995	perl select-based backend or the Event module, the POE-EV backend	1194	select-based backend or the Event module, the POE-EV backend couldn't
996	couldn't be tested because it wasn't working) shows abysmal performance	1195	be tested because it wasn't working) shows abysmal performance and
997	and memory usage: Watchers use almost 30 times as much memory as	1196	memory usage with AnyEvent: Watchers use almost 30 times as much memory
998	EV watchers, and 10 times as much memory as Event (the high memory	1197	as EV watchers, and 10 times as much memory as Event (the high memory
999	requirements are caused by requiring a session for each watcher). Watcher	1198	requirements are caused by requiring a session for each watcher). Watcher
1000	invocation speed is almost 900 times slower than with AnyEvent's pure perl	1199	invocation speed is almost 900 times slower than with AnyEvent's pure perl
		1200	implementation.
		1201
1001	implementation. The design of the POE adaptor class in AnyEvent can not	1202	The design of the POE adaptor class in AnyEvent can not really account
1002	really account for this, as session creation overhead is small compared	1203	for the performance issues, though, as session creation overhead is
1003	to execution of the state machine, which is coded pretty optimally within	1204	small compared to execution of the state machine, which is coded pretty
1004	L<AnyEvent::Impl::POE>. POE simply seems to be abysmally slow.	1205	optimally within L<AnyEvent::Impl::POE> (and while everybody agrees that
		1206	using multiple sessions is not a good approach, especially regarding
		1207	memory usage, even the author of POE could not come up with a faster
		1208	design).
1005		1209
1006	=head2 Summary	1210	=head3 Summary
1007		1211
1008	=over 4	1212	=over 4
1009		1213
1010	=item * Using EV through AnyEvent is faster than any other event loop	1214	=item * Using EV through AnyEvent is faster than any other event loop
1011	(even when used without AnyEvent), but most event loops have acceptable	1215	(even when used without AnyEvent), but most event loops have acceptable
…		…
1018	=item * You should avoid POE like the plague if you want performance or	1222	=item * You should avoid POE like the plague if you want performance or
1019	reasonable memory usage.	1223	reasonable memory usage.
1020		1224
1021	=back	1225	=back
1022		1226
		1227	=head2 BENCHMARKING THE LARGE SERVER CASE
		1228
		1229	This benchmark atcually benchmarks the event loop itself. It works by
		1230	creating a number of "servers": each server consists of a socketpair, a
		1231	timeout watcher that gets reset on activity (but never fires), and an I/O
		1232	watcher waiting for input on one side of the socket. Each time the socket
		1233	watcher reads a byte it will write that byte to a random other "server".
		1234
		1235	The effect is that there will be a lot of I/O watchers, only part of which
		1236	are active at any one point (so there is a constant number of active
		1237	fds for each loop iterstaion, but which fds these are is random). The
		1238	timeout is reset each time something is read because that reflects how
		1239	most timeouts work (and puts extra pressure on the event loops).
		1240
		1241	In this benchmark, we use 10000 socketpairs (20000 sockets), of which 100
		1242	(1%) are active. This mirrors the activity of large servers with many
		1243	connections, most of which are idle at any one point in time.
		1244
		1245	Source code for this benchmark is found as F<eg/bench2> in the AnyEvent
		1246	distribution.
		1247
		1248	=head3 Explanation of the columns
		1249
		1250	I<sockets> is the number of sockets, and twice the number of "servers" (as
		1251	each server has a read and write socket end).
		1252
		1253	I<create> is the time it takes to create a socketpair (which is
		1254	nontrivial) and two watchers: an I/O watcher and a timeout watcher.
		1255
		1256	I<request>, the most important value, is the time it takes to handle a
		1257	single "request", that is, reading the token from the pipe and forwarding
		1258	it to another server. This includes deleting the old timeout and creating
		1259	a new one that moves the timeout into the future.
		1260
		1261	=head3 Results
		1262
		1263	name sockets create request
		1264	EV 20000 69.01 11.16
		1265	Perl 20000 73.32 35.87
		1266	Event 20000 212.62 257.32
		1267	Glib 20000 651.16 1896.30
		1268	POE 20000 349.67 12317.24 uses POE::Loop::Event
		1269
		1270	=head3 Discussion
		1271
		1272	This benchmark I<does> measure scalability and overall performance of the
		1273	particular event loop.
		1274
		1275	EV is again fastest. Since it is using epoll on my system, the setup time
		1276	is relatively high, though.
		1277
		1278	Perl surprisingly comes second. It is much faster than the C-based event
		1279	loops Event and Glib.
		1280
		1281	Event suffers from high setup time as well (look at its code and you will
		1282	understand why). Callback invocation also has a high overhead compared to
		1283	the C<< $_->() for .. >>-style loop that the Perl event loop uses. Event
		1284	uses select or poll in basically all documented configurations.
		1285
		1286	Glib is hit hard by its quadratic behaviour w.r.t. many watchers. It
		1287	clearly fails to perform with many filehandles or in busy servers.
		1288
		1289	POE is still completely out of the picture, taking over 1000 times as long
		1290	as EV, and over 100 times as long as the Perl implementation, even though
		1291	it uses a C-based event loop in this case.
		1292
		1293	=head3 Summary
		1294
		1295	=over 4
		1296
		1297	=item * The pure perl implementation performs extremely well.
		1298
		1299	=item * Avoid Glib or POE in large projects where performance matters.
		1300
		1301	=back
		1302
		1303	=head2 BENCHMARKING SMALL SERVERS
		1304
		1305	While event loops should scale (and select-based ones do not...) even to
		1306	large servers, most programs we (or I :) actually write have only a few
		1307	I/O watchers.
		1308
		1309	In this benchmark, I use the same benchmark program as in the large server
		1310	case, but it uses only eight "servers", of which three are active at any
		1311	one time. This should reflect performance for a small server relatively
		1312	well.
		1313
		1314	The columns are identical to the previous table.
		1315
		1316	=head3 Results
		1317
		1318	name sockets create request
		1319	EV 16 20.00 6.54
		1320	Perl 16 25.75 12.62
		1321	Event 16 81.27 35.86
		1322	Glib 16 32.63 15.48
		1323	POE 16 261.87 276.28 uses POE::Loop::Event
		1324
		1325	=head3 Discussion
		1326
		1327	The benchmark tries to test the performance of a typical small
		1328	server. While knowing how various event loops perform is interesting, keep
		1329	in mind that their overhead in this case is usually not as important, due
		1330	to the small absolute number of watchers (that is, you need efficiency and
		1331	speed most when you have lots of watchers, not when you only have a few of
		1332	them).
		1333
		1334	EV is again fastest.
		1335
		1336	Perl again comes second. It is noticably faster than the C-based event
		1337	loops Event and Glib, although the difference is too small to really
		1338	matter.
		1339
		1340	POE also performs much better in this case, but is is still far behind the
		1341	others.
		1342
		1343	=head3 Summary
		1344
		1345	=over 4
		1346
		1347	=item * C-based event loops perform very well with small number of
		1348	watchers, as the management overhead dominates.
		1349
		1350	=back
		1351
1023		1352
1024	=head1 FORK	1353	=head1 FORK
1025		1354
1026	Most event libraries are not fork-safe. The ones who are usually are	1355	Most event libraries are not fork-safe. The ones who are usually are
1027	because they are so inefficient. Only L<EV> is fully fork-aware.	1356	because they rely on inefficient but fork-safe C<select> or C<poll>
		1357	calls. Only L<EV> is fully fork-aware.
1028		1358
1029	If you have to fork, you must either do so I<before> creating your first	1359	If you have to fork, you must either do so I<before> creating your first
1030	watcher OR you must not use AnyEvent at all in the child.	1360	watcher OR you must not use AnyEvent at all in the child.
1031		1361
1032		1362

Diff Legend

-–
+Removed lines
-+
+Added lines
-<
+Changed lines
->
+Changed lines

Comparing AnyEvent/lib/AnyEvent.pm (file contents): Revision 1.90 by root, Fri Apr 25 14:24:29 2008 UTC vs. Revision 1.105 by root, Thu May 1 12:35:54 2008 UTC

Diff Legend

Comparing AnyEvent/lib/AnyEvent.pm (file contents):
Revision 1.90 by root, Fri Apr 25 14:24:29 2008 UTC vs.
Revision 1.105 by root, Thu May 1 12:35:54 2008 UTC