MCE::Map - Parallel map model similar to the native map function
This document describes MCE::Map version 1.499_003
## Exports mce_map, mce_map_f, and mce_map_s use MCE::Map; ## Array or array_ref my @a = mce_map { $_ * $_ } 1..10000; my @b = mce_map { $_ * $_ } [ 1..10000 ]; ## File_path, glob_ref, or scalar_ref my @c = mce_map_f { chomp; do_something($_) } "/path/to/file"; my @d = mce_map_f { chomp; do_something($_) } $file_handle; my @e = mce_map_f { chomp; do_something($_) } \$scalar; ## Sequence of numbers (begin, end [, step, format]) my @f = mce_map_s { $_ * $_ } 1, 10000, 5; my @g = mce_map_s { $_ * $_ } [ 1, 10000, 5 ]; my @h = mce_map_s { $_ * $_ } { begin => 1, end => 10000, step => 5, format => undef };
This module provides a parallel map implementation via Many-core Engine. MCE incurs a small overhead due to passing of data. Therefore, a fast code block will likely run faster using the native map function in Perl. The overhead quickly diminishes as the complexity of the code block increases.
my @m1 = map { $_ * $_ } 1..1000000; ## 0.251 secs my @m2 = mce_map { $_ * $_ } 1..1000000; ## 0.525 secs
Chunking, enabled by default, greatly reduces the overhead behind the scene. The time for mce_map below also includes the time for data exchanges between the manager and worker processes. More parallelization will be seen when the code block requires more CPU code-wise.
sub calc { sqrt $_ * sqrt $_ / 1.3 * 1.5 / 3.2 * 1.07 } my @m1 = map { calc } 1..1000000; ## 0.756 secs my @m2 = mce_map { calc } 1..1000000; ## 0.623 secs
The mce_map_s funtion will provide better times, useful when input data is simply a range of numbers. Workers generate sequences mathematically among themselves without any interaction by the manager process. Two arguments are required for mce_map_s (begin, end). Step defaults to 1 if begin is smaller than end, otherwise -1.
my @m3 = mce_map_s { calc } 1, 1000000; ## 0.517 secs
Although this documentation is about MCE::Map, the MCE::Stream module can write results immediately without waiting for all chunks to complete. This is made possible by passing the reference of the array (in this case @m4 and @m5).
use MCE::Stream; sub calc { sqrt $_ * sqrt $_ / 1.3 * 1.5 / 3.2 * 1.07 } my @m4; mce_stream \@m4, sub { calc }, 1..1000000; ## Completes in 0.436 secs. That is amazing considering the ## overhead for passing data between the manager and worker. my @m5; mce_stream_s \@m5, sub { calc }, 1, 1000000; ## Completed in 0.301 secs. Like with mce_map_s, specifying a ## sequence specification turns out to be faster due to lesser ## overhead for the manager process.
The following list 5 options which may be overridden when loading the module.
use Sereal qw(encode_sereal decode_sereal); use MCE::Map max_workers => 4, ## Default 'auto' chunk_size => 100, ## Default 'auto' tmp_dir => "/path/to/app/tmp", ## $MCE::Signal::tmp_dir freeze => \&encode_sereal, ## \&Storable::freeze thaw => \&decode_sereal ## \&Storable::thaw ;
There is a simplier way to enable Sereal with MCE 1.5. The following will attempt to use Sereal if available, otherwise will default back to using Storable for serialization.
use MCE::Map Sereal => 1; ## Serialization is through Sereal if available. my @m2 = mce_map { $_ * $_ } 1..10000;
The init function takes a hash of MCE options. The gather option, if specified, will be set to undef due to being used internally by the module.
use MCE::Map; MCE::Map::init { chunk_size => 1, max_workers => 4, user_begin => sub { print "## ", MCE->wid, " started\n"; }, user_end => sub { print "## ", MCE->wid, " completed\n"; } }; my @a = mce_map { $_ * $_ } 1..100; print "\n", "@a", "\n"; -- output ## 2 started ## 1 started ## 3 started ## 4 started ## 1 completed ## 4 completed ## 2 completed ## 3 completed 1 4 9 16 25 36 49 64 81 100 121 144 169 196 225 256 289 324 361 400 441 484 529 576 625 676 729 784 841 900 961 1024 1089 1156 1225 1296 1369 1444 1521 1600 1681 1764 1849 1936 2025 2116 2209 2304 2401 2500 2601 2704 2809 2916 3025 3136 3249 3364 3481 3600 3721 3844 3969 4096 4225 4356 4489 4624 4761 4900 5041 5184 5329 5476 5625 5776 5929 6084 6241 6400 6561 6724 6889 7056 7225 7396 7569 7744 7921 8100 8281 8464 8649 8836 9025 9216 9409 9604 9801 10000
Input data can be specified using a list or passing a reference to an array.
my @a = mce_map { $_ * 2 } 1..1000; my @b = mce_map { $_ * 2 } [ 1..1000 ];
The fastest of these is the /path/to/file. Workers communicate the next offset position among themselves without any interaction by the manager process.
my @c = mce_map_f { chomp; $_ . "\r\n" } "/path/to/file"; my @d = mce_map_f { chomp; $_ . "\r\n" } $file_handle; my @e = mce_map_f { chomp; $_ . "\r\n" } \$scalar;
Sequence can be specified as a list, an array reference, or a hash reference. The functions require both begin and end values to run. Step and format are optional. The format is passed to sprintf (% can be omitted below).
my ($beg, $end, $step, $fmt) = (10, 20, 0.1, "%4.1f"); my @f = mce_map_s { $_ } $beg, $end, $step, $fmt; my @g = mce_map_s { $_ } [ $beg, $end, $step, $fmt ]; my @h = mce_map_s { $_ } { begin => $beg, end => $end, step => $step, format => $fmt };
MCE workers remain persistent as much as possible after running. Shutdown occurs when the script exits. One can manually shutdown MCE by simply calling finish after running. This resets the MCE instance.
use MCE::Map; MCE::Map::init { chunk_size => 20, max_workers => 'auto' }; my @a = mce_map { ... } 1..100; MCE::Map::finish;
MCE
Mario E. Roy, <marioeroy AT gmail DOT com>
This program is free software; you can redistribute it and/or modify it under the terms of either: the GNU General Public License as published by the Free Software Foundation; or the Artistic License.
See http://dev.perl.org/licenses/ for more information.
To install MCE, copy and paste the appropriate command in to your terminal.
cpanm
cpanm MCE
CPAN shell
perl -MCPAN -e shell install MCE
For more information on module installation, please visit the detailed CPAN module installation guide.