1MCE(3) User Contributed Perl Documentation MCE(3)
2
3
4
6 MCE - Many-Core Engine for Perl providing parallel processing
7 capabilities
8
10 This document describes MCE version 1.884
11
12 Many-Core Engine (MCE) for Perl helps enable a new level of performance
13 by maximizing all available cores.
14
16 MCE spawns a pool of workers and therefore does not fork a new process
17 per each element of data. Instead, MCE follows a bank queuing model.
18 Imagine the line being the data and bank-tellers the parallel workers.
19 MCE enhances that model by adding the ability to chunk the next n
20 elements from the input stream to the next available worker.
21
23 This is a simplistic use case of MCE running with 5 workers.
24
25 # Construction using the Core API
26
27 use MCE;
28
29 my $mce = MCE->new(
30 max_workers => 5,
31 user_func => sub {
32 my ($mce) = @_;
33 $mce->say("Hello from " . $mce->wid);
34 }
35 );
36
37 $mce->run;
38
39 # Construction using a MCE model
40
41 use MCE::Flow max_workers => 5;
42
43 mce_flow sub {
44 my ($mce) = @_;
45 MCE->say("Hello from " . MCE->wid);
46 };
47
48 The following is a demonstration for parsing a huge log file in
49 parallel.
50
51 use MCE::Loop;
52
53 MCE::Loop->init( max_workers => 8, use_slurpio => 1 );
54
55 my $pattern = 'something';
56 my $hugefile = 'very_huge.file';
57
58 my @result = mce_loop_f {
59 my ($mce, $slurp_ref, $chunk_id) = @_;
60
61 # Quickly determine if a match is found.
62 # Process the slurped chunk only if true.
63
64 if ($$slurp_ref =~ /$pattern/m) {
65 my @matches;
66
67 # The following is fast on Unix, but performance degrades
68 # drastically on Windows beyond 4 workers.
69
70 open my $MEM_FH, '<', $slurp_ref;
71 binmode $MEM_FH, ':raw';
72 while (<$MEM_FH>) { push @matches, $_ if (/$pattern/); }
73 close $MEM_FH;
74
75 # Therefore, use the following construction on Windows.
76
77 while ( $$slurp_ref =~ /([^\n]+\n)/mg ) {
78 my $line = $1; # save $1 to not lose the value
79 push @matches, $line if ($line =~ /$pattern/);
80 }
81
82 # Gather matched lines.
83
84 MCE->gather(@matches);
85 }
86
87 } $hugefile;
88
89 print join('', @result);
90
91 The next demonstration loops through a sequence of numbers with
92 MCE::Flow.
93
94 use MCE::Flow;
95
96 my $N = shift || 4_000_000;
97
98 sub compute_pi {
99 my ( $beg_seq, $end_seq ) = @_;
100 my ( $pi, $t ) = ( 0.0 );
101
102 foreach my $i ( $beg_seq .. $end_seq ) {
103 $t = ( $i + 0.5 ) / $N;
104 $pi += 4.0 / ( 1.0 + $t * $t );
105 }
106
107 MCE->gather( $pi );
108 }
109
110 # Compute bounds only, workers receive [ begin, end ] values
111
112 MCE::Flow->init(
113 chunk_size => 200_000,
114 max_workers => 8,
115 bounds_only => 1
116 );
117
118 my @ret = mce_flow_s sub {
119 compute_pi( $_->[0], $_->[1] );
120 }, 0, $N - 1;
121
122 my $pi = 0.0; $pi += $_ for @ret;
123
124 printf "pi = %0.13f\n", $pi / $N; # 3.1415926535898
125
127 Four modules make up the core engine for MCE.
128
129 MCE::Core
130 This is the POD documentation describing the core Many-Core Engine
131 (MCE) API. Go here for help with the various MCE options. See also,
132 MCE::Examples for additional demonstrations.
133
134 MCE::Mutex
135 Provides a simple semaphore implementation supporting threads and
136 processes. Two implementations are provided; one via pipes or
137 socket depending on the platform and the other using Fcntl.
138
139 MCE::Signal
140 Provides signal handling, temporary directory creation, and cleanup
141 for MCE.
142
143 MCE::Util
144 Provides utility functions for MCE.
145
147 There are 5 add-on modules for use with MCE.
148
149 MCE::Candy
150 Provides a collection of sugar methods and output iterators for
151 preserving output order.
152
153 MCE::Channel
154 Introduced in MCE 1.839, provides queue-like and two-way
155 communication capability. Three implementations "Simple", "Mutex",
156 and "Threads" are provided. "Simple" does not involve locking
157 whereas "Mutex" and "Threads" do locking transparently using
158 "MCE::Mutex" and "threads" respectively.
159
160 MCE::Child
161 Also introduced in MCE 1.839, provides a threads-like
162 parallelization module that is compatible with Perl 5.8. It is a
163 fork of MCE::Hobo. The difference is using a common "MCE::Channel"
164 object when yielding and joining.
165
166 MCE::Queue
167 Provides a hybrid queuing implementation for MCE supporting normal
168 queues and priority queues from a single module. MCE::Queue
169 exchanges data via the core engine to enable queuing to work for
170 both children (spawned from fork) and threads.
171
172 MCE::Relay
173 Provides workers the ability to receive and pass information orderly
174 with zero involvement by the manager process. This module is loaded
175 automatically by MCE when specifying the "init_relay" MCE option.
176
178 The MCE models are sugar syntax on top of the MCE::Core API. Two MCE
179 options (chunk_size and max_workers) are configured automatically.
180 Moreover, spawning workers and later shutdown occur transparently
181 behind the scene.
182
183 Choosing a MCE Model largely depends on the application. It all boils
184 down to how much automation you need MCE to handle transparently. Or if
185 you prefer, constructing the MCE object and running using the core MCE
186 API is fine too.
187
188 MCE::Grep
189 Provides a parallel grep implementation similar to the native grep
190 function.
191
192 MCE::Map
193 Provides a parallel map implementation similar to the native map
194 function.
195
196 MCE::Loop
197 Provides a parallel for loop implementation.
198
199 MCE::Flow
200 Like "MCE::Loop", but with support for multiple pools of workers.
201 The pool of workers are configured transparently via the MCE
202 "user_tasks" option.
203
204 MCE::Step
205 Like "MCE::Flow", but adds a "MCE::Queue" object between each pool
206 of workers. This model, introduced in 1.506, allows one to pass data
207 forward (left to right) from one sub-task into another with little
208 effort.
209
210 MCE::Stream
211 This provides an efficient parallel implementation for chaining
212 multiple maps and greps transparently. Like "MCE::Flow" and
213 "MCE::Step", it too supports multiple pools of workers. The
214 distinction is that "MCE::Stream" passes data from right to left and
215 done for you transparently.
216
218 Miscellaneous additions included with the distribution.
219
220 MCE::Examples
221 Describes various demonstrations for MCE including a Monte Carlo
222 simulation.
223
224 MCE::Subs
225 Exports functions mapped directly to MCE methods; e.g. mce_wid. The
226 module allows 3 options; :manager, :worker, and :getter.
227
229 Perl 5.8.0 or later.
230
232 The source and examples are hosted at GitHub.
233
234 • <https://github.com/marioroy/mce-perl>
235
236 • <https://github.com/marioroy/mce-examples>
237
239 Refer to the MCE::Core documentation where the API is described.
240
241 "MCE::Shared" provides data sharing capabilities for "MCE". It includes
242 "MCE::Hobo" for running code asynchronously with the IPC handled by the
243 shared-manager process.
244
245 • MCE::Shared
246
247 • MCE::Hobo
248
250 Mario E. Roy, <marioeroy AT gmail DOT com>
251
253 Copyright (C) 2012-2023 by Mario E. Roy
254
255 MCE is released under the same license as Perl.
256
257 See <https://dev.perl.org/licenses/> for more information.
258
259
260
261perl v5.36.0 2023-01-20 MCE(3)