ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/AnyEvent/lib/AnyEvent.pm
(Generate patch)

Comparing AnyEvent/lib/AnyEvent.pm (file contents):
Revision 1.277 by root, Sun Aug 9 13:27:23 2009 UTC vs.
Revision 1.278 by root, Sun Aug 9 15:09:28 2009 UTC

… …
1288 # we assume CLOEXEC is already set by perl in all important cases 1288 # we assume CLOEXEC is already set by perl in all important cases
1289 1289
1290 ($fh2, $rw) 1290 ($fh2, $rw)
1291} 1291}
1292 1292
1293############################################################################# 1293=head1 SIMPLIFIED AE API
1294# "new" API, currently only emulation of it 1294
1295############################################################################# 1295Starting with version 5.0, AnyEvent officially supports a second, much
1296simpler, API that is designed to reduce the calling, typing and memory
1297overhead.
1298
1299See the L<AE> manpage for details.
1300
1301=cut
1296 1302
1297package AE; 1303package AE;
1298 1304
1299our $VERSION = $AnyEvent::VERSION; 1305our $VERSION = $AnyEvent::VERSION;
1300 1306
… …
2055through AnyEvent. The benchmark creates a lot of timers (with a zero 2061through AnyEvent. The benchmark creates a lot of timers (with a zero
2056timeout) and I/O watchers (watching STDOUT, a pty, to become writable, 2062timeout) and I/O watchers (watching STDOUT, a pty, to become writable,
2057which it is), lets them fire exactly once and destroys them again. 2063which it is), lets them fire exactly once and destroys them again.
2058 2064
2059Source code for this benchmark is found as F<eg/bench> in the AnyEvent 2065Source code for this benchmark is found as F<eg/bench> in the AnyEvent
2060distribution. 2066distribution. It uses the L<AE> interface, which makes a real difference
2067for the EV and Perl backends only.
2061 2068
2062=head3 Explanation of the columns 2069=head3 Explanation of the columns
2063 2070
2064I<watcher> is the number of event watchers created/destroyed. Since 2071I<watcher> is the number of event watchers created/destroyed. Since
2065different event models feature vastly different performances, each event 2072different event models feature vastly different performances, each event
… …
2086watcher. 2093watcher.
2087 2094
2088=head3 Results 2095=head3 Results
2089 2096
2090 name watchers bytes create invoke destroy comment 2097 name watchers bytes create invoke destroy comment
2091 EV/EV 400000 224 0.47 0.35 0.27 EV native interface 2098 EV/EV 100000 223 0.47 0.43 0.27 EV native interface
2092 EV/Any 100000 224 2.88 0.34 0.27 EV + AnyEvent watchers 2099 EV/Any 100000 223 0.48 0.42 0.26 EV + AnyEvent watchers
2093 CoroEV/Any 100000 224 2.85 0.35 0.28 coroutines + Coro::Signal 2100 Coro::EV/Any 100000 223 0.47 0.42 0.26 coroutines + Coro::Signal
2094 Perl/Any 100000 452 4.13 0.73 0.95 pure perl implementation 2101 Perl/Any 100000 431 2.70 0.74 0.92 pure perl implementation
2095 Event/Event 16000 517 32.20 31.80 0.81 Event native interface 2102 Event/Event 16000 516 31.16 31.84 0.82 Event native interface
2096 Event/Any 16000 590 35.85 31.55 1.06 Event + AnyEvent watchers 2103 Event/Any 16000 1203 42.61 34.79 1.80 Event + AnyEvent watchers
2097 IOAsync/Any 16000 989 38.10 32.77 11.13 via IO::Async::Loop::IO_Poll 2104 IOAsync/Any 16000 1911 41.92 27.45 16.81 via IO::Async::Loop::IO_Poll
2098 IOAsync/Any 16000 990 37.59 29.50 10.61 via IO::Async::Loop::Epoll 2105 IOAsync/Any 16000 1726 40.69 26.37 15.25 via IO::Async::Loop::Epoll
2099 Glib/Any 16000 1357 102.33 12.31 51.00 quadratic behaviour 2106 Glib/Any 16000 1118 89.00 12.57 51.17 quadratic behaviour
2100 Tk/Any 2000 1860 27.20 66.31 14.00 SEGV with >> 2000 watchers 2107 Tk/Any 2000 1346 20.96 10.75 8.00 SEGV with >> 2000 watchers
2101 POE/Event 2000 6328 109.99 751.67 14.02 via POE::Loop::Event 2108 POE/Any 2000 6951 108.97 795.32 14.24 via POE::Loop::Event
2102 POE/Select 2000 6027 94.54 809.13 579.80 via POE::Loop::Select 2109 POE/Any 2000 6648 94.79 774.40 575.51 via POE::Loop::Select
2103 2110
2104=head3 Discussion 2111=head3 Discussion
2105 2112
2106The benchmark does I<not> measure scalability of the event loop very 2113The benchmark does I<not> measure scalability of the event loop very
2107well. For example, a select-based event loop (such as the pure perl one) 2114well. For example, a select-based event loop (such as the pure perl one)
… …
2119benchmark machine, handling an event takes roughly 1600 CPU cycles with 2126benchmark machine, handling an event takes roughly 1600 CPU cycles with
2120EV, 3100 CPU cycles with AnyEvent's pure perl loop and almost 3000000 CPU 2127EV, 3100 CPU cycles with AnyEvent's pure perl loop and almost 3000000 CPU
2121cycles with POE. 2128cycles with POE.
2122 2129
2123C<EV> is the sole leader regarding speed and memory use, which are both 2130C<EV> is the sole leader regarding speed and memory use, which are both
2124maximal/minimal, respectively. Even when going through AnyEvent, it uses 2131maximal/minimal, respectively. When using the L<AE> API there is zero
2132overhead (when going through the AnyEvent API create is about 5-6 times
2133slower, with other times being equal, so still uses far less memory than
2125far less memory than any other event loop and is still faster than Event 2134any other event loop and is still faster than Event natively).
2126natively.
2127 2135
2128The pure perl implementation is hit in a few sweet spots (both the 2136The pure perl implementation is hit in a few sweet spots (both the
2129constant timeout and the use of a single fd hit optimisations in the perl 2137constant timeout and the use of a single fd hit optimisations in the perl
2130interpreter and the backend itself). Nevertheless this shows that it 2138interpreter and the backend itself). Nevertheless this shows that it
2131adds very little overhead in itself. Like any select-based backend its 2139adds very little overhead in itself. Like any select-based backend its
… …
2205In this benchmark, we use 10000 socket pairs (20000 sockets), of which 100 2213In this benchmark, we use 10000 socket pairs (20000 sockets), of which 100
2206(1%) are active. This mirrors the activity of large servers with many 2214(1%) are active. This mirrors the activity of large servers with many
2207connections, most of which are idle at any one point in time. 2215connections, most of which are idle at any one point in time.
2208 2216
2209Source code for this benchmark is found as F<eg/bench2> in the AnyEvent 2217Source code for this benchmark is found as F<eg/bench2> in the AnyEvent
2210distribution. 2218distribution. It uses the L<AE> interface, which makes a real difference
2219for the EV and Perl backends only.
2211 2220
2212=head3 Explanation of the columns 2221=head3 Explanation of the columns
2213 2222
2214I<sockets> is the number of sockets, and twice the number of "servers" (as 2223I<sockets> is the number of sockets, and twice the number of "servers" (as
2215each server has a read and write socket end). 2224each server has a read and write socket end).
… …
2223a new one that moves the timeout into the future. 2232a new one that moves the timeout into the future.
2224 2233
2225=head3 Results 2234=head3 Results
2226 2235
2227 name sockets create request 2236 name sockets create request
2228 EV 20000 69.01 11.16 2237 EV 20000 62.66 7.99
2229 Perl 20000 73.32 35.87 2238 Perl 20000 68.32 32.64
2230 IOAsync 20000 157.00 98.14 epoll 2239 IOAsync 20000 174.06 101.15 epoll
2231 IOAsync 20000 159.31 616.06 poll 2240 IOAsync 20000 174.67 610.84 poll
2232 Event 20000 212.62 257.32 2241 Event 20000 202.69 242.91
2233 Glib 20000 651.16 1896.30 2242 Glib 20000 557.01 1689.52
2234 POE 20000 349.67 12317.24 uses POE::Loop::Event 2243 POE 20000 341.54 12086.32 uses POE::Loop::Event
2235 2244
2236=head3 Discussion 2245=head3 Discussion
2237 2246
2238This benchmark I<does> measure scalability and overall performance of the 2247This benchmark I<does> measure scalability and overall performance of the
2239particular event loop. 2248particular event loop.

Diff Legend

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