… | |
… | |
136 | |
136 | |
137 | Note that C<my $w; $w => combination. This is necessary because in Perl, |
137 | Note that C<my $w; $w => combination. This is necessary because in Perl, |
138 | my variables are only visible after the statement in which they are |
138 | my variables are only visible after the statement in which they are |
139 | declared. |
139 | declared. |
140 | |
140 | |
141 | =head2 IO WATCHERS |
141 | =head2 I/O WATCHERS |
142 | |
142 | |
143 | You can create an I/O watcher by calling the C<< AnyEvent->io >> method |
143 | You can create an I/O watcher by calling the C<< AnyEvent->io >> method |
144 | with the following mandatory key-value pairs as arguments: |
144 | with the following mandatory key-value pairs as arguments: |
145 | |
145 | |
146 | C<fh> the Perl I<file handle> (I<not> file descriptor) to watch for |
146 | C<fh> the Perl I<file handle> (I<not> file descriptor) to watch for |
… | |
… | |
706 | |
706 | |
707 | =back |
707 | =back |
708 | |
708 | |
709 | =head1 EXAMPLE PROGRAM |
709 | =head1 EXAMPLE PROGRAM |
710 | |
710 | |
711 | The following program uses an IO watcher to read data from STDIN, a timer |
711 | The following program uses an I/O watcher to read data from STDIN, a timer |
712 | to display a message once per second, and a condition variable to quit the |
712 | to display a message once per second, and a condition variable to quit the |
713 | program when the user enters quit: |
713 | program when the user enters quit: |
714 | |
714 | |
715 | use AnyEvent; |
715 | use AnyEvent; |
716 | |
716 | |
… | |
… | |
863 | $quit->wait; |
863 | $quit->wait; |
864 | |
864 | |
865 | |
865 | |
866 | =head1 BENCHMARK |
866 | =head1 BENCHMARK |
867 | |
867 | |
868 | To give you an idea of the performance an doverheads that AnyEvent adds |
868 | To give you an idea of the performance and overheads that AnyEvent adds |
869 | over the backends, here is a benchmark of various supported backends. The |
869 | over the event loops themselves (and to give you an impression of the |
870 | benchmark creates a lot of timers (with zero timeout) and io events |
870 | speed of various event loops), here is a benchmark of various supported |
871 | (watching STDOUT, a pty, to become writable). |
871 | event models natively and with anyevent. The benchmark creates a lot of |
|
|
872 | timers (with a zero timeout) and I/O watchers (watching STDOUT, a pty, to |
|
|
873 | become writable, which it is), lets them fire exactly once and destroys |
|
|
874 | them again. |
872 | |
875 | |
873 | Explanation of the fields: |
876 | Rewriting the benchmark to use many different sockets instead of using |
|
|
877 | the same filehandle for all I/O watchers results in a much longer runtime |
|
|
878 | (socket creation is expensive), but qualitatively the same figures, so it |
|
|
879 | was not used. |
874 | |
880 | |
|
|
881 | =head2 Explanation of the columns |
|
|
882 | |
875 | I<watcher> is the number of event watchers created/destroyed. Sicne |
883 | I<watcher> is the number of event watchers created/destroyed. Since |
876 | different event models have vastly different performance each backend was |
884 | different event models feature vastly different performances, each event |
877 | handed a number of watchers so that overall runtime is acceptable and |
885 | loop was given a number of watchers so that overall runtime is acceptable |
878 | similar to all backends (and keep them from crashing). |
886 | and similar between tested event loop (and keep them from crashing): Glib |
|
|
887 | would probably take thousands of years if asked to process the same number |
|
|
888 | of watchers as EV in this benchmark. |
879 | |
889 | |
880 | I<bytes> is the number of bytes (as measured by resident set size) used by |
890 | I<bytes> is the number of bytes (as measured by the resident set size, |
881 | each watcher. |
891 | RSS) consumed by each watcher. This method of measuring captures both C |
|
|
892 | and Perl-based overheads. |
882 | |
893 | |
883 | I<create> is the time, in microseconds, to create a single watcher. |
894 | I<create> is the time, in microseconds (millionths of seconds), that it |
|
|
895 | takes to create a single watcher. The callback is a closure shared between |
|
|
896 | all watchers, to avoid adding memory overhead. That means closure creation |
|
|
897 | and memory usage is not included in the figures. |
884 | |
898 | |
885 | I<invoke> is the time, in microseconds, used to invoke a simple callback |
899 | I<invoke> is the time, in microseconds, used to invoke a simple |
886 | that simply counts down. |
900 | callback. The callback simply counts down a Perl variable and after it was |
|
|
901 | invoked "watcher" times, it would C<< ->broadcast >> a condvar once to |
|
|
902 | signal the end of this phase. |
887 | |
903 | |
888 | I<destroy> is the time, in microseconds, to destroy a single watcher. |
904 | I<destroy> is the time, in microseconds, that it takes to destroy a single |
|
|
905 | watcher. |
889 | |
906 | |
|
|
907 | =head2 Results |
|
|
908 | |
890 | name watcher bytes create invoke destroy comment |
909 | name watchers bytes create invoke destroy comment |
891 | EV/EV 400000 244 0.56 0.46 0.31 EV native interface |
910 | EV/EV 400000 244 0.56 0.46 0.31 EV native interface |
892 | EV/Any 100000 610 3.52 0.91 0.75 |
911 | EV/Any 100000 610 3.52 0.91 0.75 EV + AnyEvent watchers |
893 | CoroEV/Any 100000 610 3.49 0.92 0.75 coroutines + Coro::Signal |
912 | CoroEV/Any 100000 610 3.49 0.92 0.75 coroutines + Coro::Signal |
894 | Perl/Any 10000 654 4.64 1.22 0.77 pure perl implementation |
913 | Perl/Any 100000 513 4.91 0.92 1.15 pure perl implementation |
895 | Event/Event 10000 523 28.05 21.38 5.22 Event native interface |
914 | Event/Event 16000 523 28.05 21.38 0.86 Event native interface |
896 | Event/Any 10000 943 34.43 20.48 1.39 |
915 | Event/Any 16000 943 34.43 20.48 1.39 Event + AnyEvent watchers |
897 | Glib/Any 16000 1357 96.99 12.55 55.51 quadratic behaviour |
916 | Glib/Any 16000 1357 96.99 12.55 55.51 quadratic behaviour |
898 | Tk/Any 2000 1855 27.01 66.61 14.03 SEGV with >> 2000 watchers |
917 | Tk/Any 2000 1855 27.01 66.61 14.03 SEGV with >> 2000 watchers |
|
|
918 | POE/Event 2000 6644 108.15 768.19 14.33 via POE::Loop::Event |
899 | POE/Select 2000 6343 94.69 807.65 562.69 POE::Loop::Select |
919 | POE/Select 2000 6343 94.69 807.65 562.69 via POE::Loop::Select |
900 | POE/Event 2000 6644 108.15 768.19 14.33 POE::Loop::Event |
|
|
901 | |
920 | |
902 | Discussion: The benchmark does I<not> bench scalability of the |
921 | =head2 Discussion |
|
|
922 | |
|
|
923 | The benchmark does I<not> measure scalability of the event loop very |
903 | backend. For example a select-based backend (such as the pureperl one) can |
924 | well. For example, a select-based event loop (such as the pure perl one) |
904 | never compete with a backend using epoll. In this benchmark, only a single |
925 | can never compete with an event loop that uses epoll when the number of |
905 | filehandle is used. |
926 | file descriptors grows high. In this benchmark, only a single filehandle |
|
|
927 | is used (although some of the AnyEvent adaptors dup() its file descriptor |
|
|
928 | to worka round bugs). |
906 | |
929 | |
907 | EV is the sole leader regarding speed and memory use, which are both |
930 | C<EV> is the sole leader regarding speed and memory use, which are both |
908 | maximal/minimal. Even when going through AnyEvent, there is only one event |
931 | maximal/minimal, respectively. Even when going through AnyEvent, there are |
909 | loop that uses less memory (the Event module natively), and no faster |
932 | only two event loops that use slightly less memory (the C<Event> module |
910 | event model. |
933 | natively and the pure perl backend), and no faster event models, not even |
|
|
934 | C<Event> natively. |
911 | |
935 | |
912 | The pure perl implementation is hit in a few sweet spots (both the |
936 | The pure perl implementation is hit in a few sweet spots (both the |
913 | zero timeout and the use of a single fd hit optimisations in the perl |
937 | zero timeout and the use of a single fd hit optimisations in the perl |
914 | interpreter and the backend itself), but it shows that it adds very little |
938 | interpreter and the backend itself, and all watchers become ready at the |
|
|
939 | same time). Nevertheless this shows that it adds very little overhead in |
915 | overhead in itself. Like any select-based backend it's performance becomes |
940 | itself. Like any select-based backend its performance becomes really bad |
916 | really bad with lots of file descriptors. |
941 | with lots of file descriptors (and few of them active), of course, but |
|
|
942 | this was not subject of this benchmark. |
917 | |
943 | |
918 | The Event module has a relatively high setup and callback invocation cost, |
944 | The C<Event> module has a relatively high setup and callback invocation cost, |
919 | but overall scores on the third place. |
945 | but overall scores on the third place. |
920 | |
946 | |
921 | Glib has a little higher memory cost, a bit fster callback invocation and |
947 | C<Glib>'s memory usage is quite a bit bit higher, but it features a |
922 | has a similar speed as Event. |
948 | faster callback invocation and overall ends up in the same class as |
|
|
949 | C<Event>. However, Glib scales extremely badly, doubling the number of |
|
|
950 | watchers increases the processing time by more than a factor of four, |
|
|
951 | making it completely unusable when using larger numbers of watchers |
|
|
952 | (note that only a single file descriptor was used in the benchmark, so |
|
|
953 | inefficiencies of C<poll> do not account for this). |
923 | |
954 | |
924 | The Tk backend works relatively well, the fact that it crashes with |
955 | The C<Tk> adaptor works relatively well. The fact that it crashes with |
925 | more than 2000 watchers is a big setback, however, as correctness takes |
956 | more than 2000 watchers is a big setback, however, as correctness takes |
926 | precedence over speed. |
957 | precedence over speed. Nevertheless, its performance is surprising, as the |
|
|
958 | file descriptor is dup()ed for each watcher. This shows that the dup() |
|
|
959 | employed by some adaptors is not a big performance issue (it does incur a |
|
|
960 | hidden memory cost inside the kernel, though, that is not reflected in the |
|
|
961 | figures above). |
927 | |
962 | |
928 | POE, regardless of backend (wether it's pure perl select backend or the |
963 | C<POE>, regardless of underlying event loop (wether using its pure perl |
929 | Event backend) shows abysmal performance and memory usage: Watchers use |
964 | select-based backend or the Event module) shows abysmal performance and |
930 | almost 30 times as much memory as EV watchers, and 10 times as much memory |
965 | memory usage: Watchers use almost 30 times as much memory as EV watchers, |
931 | as both Event or EV via AnyEvent. |
966 | and 10 times as much memory as both Event or EV via AnyEvent. Watcher |
|
|
967 | invocation is almost 900 times slower than with AnyEvent's pure perl |
|
|
968 | implementation. The design of the POE adaptor class in AnyEvent can not |
|
|
969 | really account for this, as session creation overhead is small compared |
|
|
970 | to execution of the state machine, which is coded pretty optimally within |
|
|
971 | L<AnyEvent::Impl::POE>. POE simply seems to be abysmally slow. |
932 | |
972 | |
|
|
973 | =head2 Summary |
|
|
974 | |
933 | Summary: using EV through AnyEvent is faster than any other event |
975 | Using EV through AnyEvent is faster than any other event loop, but most |
934 | loop. The overhead AnyEvent adds can be very small, and you should avoid |
976 | event loops have acceptable performance with or without AnyEvent. |
935 | POE like the plague if you want performance or reasonable memory usage. |
977 | |
|
|
978 | The overhead AnyEvent adds is usually much smaller than the overhead of |
|
|
979 | the actual event loop, only with extremely fast event loops such as the EV |
|
|
980 | adds AnyEvent significant overhead. |
|
|
981 | |
|
|
982 | And you should simply avoid POE like the plague if you want performance or |
|
|
983 | reasonable memory usage. |
936 | |
984 | |
937 | |
985 | |
938 | =head1 FORK |
986 | =head1 FORK |
939 | |
987 | |
940 | Most event libraries are not fork-safe. The ones who are usually are |
988 | Most event libraries are not fork-safe. The ones who are usually are |