Beefy Boxes and Bandwidth Generously Provided by pair Networks
Do you know where your variables are?
 
PerlMonks  

Re^2: Problem in RAM usage while threading the program

by marioroy (Parson)
on Dec 20, 2019 at 08:50 UTC ( #11110419=note: print w/replies, xml ) Need Help??


in reply to Re: Problem in RAM usage while threading the program
in thread Problem in RAM usage while threading the program

This is a continuation of Re: Problem in RAM usage while threading the program.

Update 1: Added parallel chunking demonstration.
Update 2: Inline::C demonstrations can be found here.

Part Three create the Kyoto Cabinet DB

This is where the fun begins. The # after the file name is how to pass options. See the perldoc for more info. Notice the split line. There is no reason to split the entire sequence header line.

# https://www.perlmonks.org/?node_id=11110379 # usage: perl create_db.pl use strict; use warnings; use KyotoCabinet; unlink 'db.kch'; # construct the database object my $db = KyotoCabinet::DB->new(); # open the database if (! $db->open('db.kch#msiz=128m', $db->OWRITER | $db->OCREATE)) { die "open error (db): ", $db->error; } open my $fh, '<', 'NR.fasta' or die "open error: $!\n"; my $nrid = ''; my $seq = ''; while ( <$fh> ) { chomp; $_ =~ s/\r$//g; if ( /^>/ ) { $db->set($nrid, $seq) if $seq; $nrid = ( split /\s/, $_, 2 )[0]; $seq = ''; } else { $seq .= $_; } } $db->set($nrid, $seq) if $seq; close $fh; $db->close;

Part Four search DB serial code

One might like to measure the difference between running serially and parallel. This makes use of cursor in Kyoto Cabinet. The substr is for stripping the leading comma when displaying the value to STDOUT.

# https://www.perlmonks.org/?node_id=11110379 # usage: perl search_db.pl > Outfile.txt use strict; use warnings; use KyotoCabinet; # construct the database object my $db = KyotoCabinet::DB->new(); # open the database if (! $db->open('db.kch#msiz=128m', $db->OREADER | $db->ONOLOCK)) { die "open error (db): ", $db->error; } my $cur = $db->cursor; open my $fh, '<', 'peptides.txt' or die "open error: $!\n"; while ( my $pep = <$fh> ) { my $ids = ''; chomp $pep; $cur->jump; # first record while ( my ($key, $val) = $cur->get(1) ) { $ids .= ",$key" if index($val, $pep) >= 0; } print "$pep\t", substr($ids, 1), "\n" if $ids; } close $fh; $cur->disable; $db->close;

Part Five search DB parallel code

Here it is. This is the parallel code. It's practically the serial code above wrapped into MCE. Each worker opens a handle to the DB. This is done inside the user_begin block. Likewise, closing is done inside the user_end block. MCE::Relay is helpful for ensuring workers write serially to STDOUT. Thus, no reason to have workers send output to the manager process or store in a shared hash (i.e. MCE::Shared).

# https://www.perlmonks.org/?node_id=11110379 # usage: perl search_db_mce.pl > Outfile.txt use strict; use warnings; use KyotoCabinet; use MCE; # construct the database object my $db = KyotoCabinet::DB->new(); my $cur; my $mce = MCE->new( max_workers => MCE::Util::get_ncpu(), chunk_size => 1, init_relay => 1, user_begin => sub { # open the database if (! $db->open('db.kch#msiz=128m', $db->OREADER | $db->ONOLOC +K)) { die "open error (db): ", $db->error; } $cur = $db->cursor; }, user_end => sub { # close the database $cur->disable; $db->close; }, user_func => sub { my $pep = $_; chomp $pep; my $ids = ''; $cur->jump; # first record while ( my ($key, $val) = $cur->get(1) ) { $ids .= ",$key" if index($val, $pep) >= 0; } # output serially, one worker at a time MCE::relay { print "$pep\t", substr($ids, 1), "\n" if $ids; }; } ); $mce->process('peptides.txt'); $mce->shutdown;

Part Six search DB parallel code - chunking

Chunking helps reduce time by another 25% ~ 30% for this use case. This is beneficial on machines with many cores.

# https://www.perlmonks.org/?node_id=11110379 # usage: perl search_db_chunk.pl > Outfile.txt use strict; use warnings; use KyotoCabinet; use MCE; # construct the database object my $db = KyotoCabinet::DB->new(); my $cur; my $mce = MCE->new( max_workers => MCE::Util::get_ncpu(), chunk_size => 7, init_relay => 1, user_begin => sub { # open the database if (! $db->open('db.kch#msiz=128m', $db->OREADER | $db->ONOLOC +K)) { die "open error (db): ", $db->error; } $cur = $db->cursor; }, user_end => sub { # close the database $cur->disable; $db->close; }, user_func => sub { my ( $mce, $chunk_ref, $chunk_id ) = @_; my %ret; chomp @$chunk_ref; $cur->jump; # first record while ( my ($key, $val) = $cur->get(1) ) { for ( @$chunk_ref ) { $ret{$_} .= ",$key" if index($val, $_) >= 0; } } # output serially, one worker at a time MCE::relay { for ( @$chunk_ref ) { print "$_\t", substr($ret{$_}, 1), "\n" if $ret{$_}; } }; } ); $mce->process('peptides.txt'); $mce->shutdown;

Outfile.txt

GAAGGACTGGGACCA >NR_000001,>NR_006611 AGGCTGCGGCAGGAC >NR_062102 GTGAGCCGGGCAGAG >NR_089584 AGGGGGGGTTGCTGA >NR_036454,>NR_068535,>NR_097889 CTGACATGCGGCGCA >NR_087289 GTGCATCGATGGCCG >NR_005535 GGGGTCAAGCGAACC >NR_076289,>NR_087856 TGAGACGGCGAACCT >NR_064242 AGCGACAAAGGAAAC >NR_045865 AGGTGCAACCATGGA >NR_046602,>NR_056869 GAGTAAACCGCGCGA >NR_093455 AACGACTGAACAGCG >NR_070693 ACGCGTAATTCGATA >NR_080086 GATGAGCGGAGCACT >NR_070118 CGTAGCGAAACCGAG >NR_092384,>NR_098291 GGGGGGGAGGTCCGA >NR_021671,>NR_036907,>NR_080961 AGGGAGGGGGGTTGT >NR_026207 ATGGGGCAGACGCGA >NR_072314

Regards, Mario

Replies are listed 'Best First'.
Re^3: Problem in RAM usage while threading the program
by marioroy (Parson) on Dec 23, 2019 at 01:35 UTC

    This is a continuation of Re: Problem in RAM usage while threading the program.

    Update: Added Inline::CPP code.

    Part Seven Inline::C and Inline::CPP demonstrations.

    There was one more thing to try. Of all things, Inline::C / CPP on Linux to see what improvements that would bring.

    Inline::C

    # https://fallabs.com/kyotocabinet/api/ use Inline C => config => inc => '-I/usr/local/include'; use Inline C => config => libs => '-L/usr/local/lib -lkyotocabinet'; use Inline C => <<'EOC'; #include <string.h> #include <kclangc.h> KCDB *db; KCCUR *cur; // open the database int open_db(char* file) { db = kcdbnew(); if (!kcdbopen(db, file, KCOREADER | KCONOLOCK)) return 0; cur = kcdbcursor(db); return 1; } // close the database int close_db() { if (cur) kccurdel(cur); if (!kcdbclose(db)) return 0; return 1; } // search records containing substring SV * search_db(char* substr) { AV *ret = newAV(); char *kbuf, *vbuf; size_t ksiz, vsiz; const char *cvbuf; kccurjump(cur); while ((kbuf = kccurget(cur, &ksiz, &cvbuf, &vsiz, 1)) != NULL) { if (strstr(cvbuf, substr) != NULL) { av_push(ret, newSVpvn(kbuf, ksiz)); } kcfree(kbuf); } return newRV_noinc((SV *) ret); } EOC

    Inline::CPP

    # https://fallabs.com/kyotocabinet/api/ use Inline CPP => config => inc => '-I/usr/local/include'; use Inline CPP => config => libs => '-L/usr/local/lib -lkyotocabinet'; use Inline CPP => <<'EOCPP'; #undef do_open #undef do_close #include <string.h> #include <kcpolydb.h> using namespace std; using namespace kyotocabinet; PolyDB db; DB::Cursor *cur; // open the database int open_db(char* file) { if (!db.open(file, PolyDB::OREADER | PolyDB::ONOLOCK)) return 0; cur = db.cursor(); return 1; } // close the database int close_db() { if (cur) delete cur; if (!db.close()) return 0; return 1; } // search records containing substring SV * search_db(char* substr) { AV *ret = newAV(); string ckey, cvalue; cur->jump(); while (cur->get(&ckey, &cvalue, true)) { if (strstr(cvalue.c_str(), substr) != NULL) { av_push(ret, newSVpvn(ckey.c_str(), ckey.length())); } } return newRV_noinc((SV *) ret); } EOCPP

    Serial

    # https://www.perlmonks.org/?node_id=11110379 # usage: perl search_db_inline_c.pl > Outfile.txt use strict; use warnings; # insert the Inline::C or Inline::CPP code here open_db('db.kch#msiz=128m') or die "db.kch: open error\n"; open my $fh, '<', 'peptides.txt' or die "open error: $!\n"; while ( my $pep = <$fh> ) { chomp $pep; my $ids = search_db($pep); print "$pep\t", join(',', @$ids), "\n" if @$ids; } close $fh; close_db();

    Parallel

    # https://www.perlmonks.org/?node_id=11110379 # usage: perl search_db_inline_c_mce.pl > Outfile.txt use strict; use warnings; # insert the Inline::C or Inline::CPP code here use MCE; my $mce = MCE->new( max_workers => MCE::Util::get_ncpu(), chunk_size => 1, init_relay => 1, user_begin => sub { open_db('db.kch#msiz=128m') or die "db.kch: open error\n"; }, user_end => sub { close_db(); }, user_func => sub { my $pep = $_; chomp $pep; my $ids = search_db($pep); # output serially, one worker at a time MCE::relay { print "$pep\t", join(',', @$ids), "\n" if @$ids; }; } ); $mce->process('peptides.txt'); $mce->shutdown;

    Outfile.txt

    GAAGGACTGGGACCA >NR_000001,>NR_006611 AGGCTGCGGCAGGAC >NR_062102 GTGAGCCGGGCAGAG >NR_089584 AGGGGGGGTTGCTGA >NR_036454,>NR_068535,>NR_097889 CTGACATGCGGCGCA >NR_087289 GTGCATCGATGGCCG >NR_005535 GGGGTCAAGCGAACC >NR_076289,>NR_087856 TGAGACGGCGAACCT >NR_064242 AGCGACAAAGGAAAC >NR_045865 AGGTGCAACCATGGA >NR_046602,>NR_056869 GAGTAAACCGCGCGA >NR_093455 AACGACTGAACAGCG >NR_070693 ACGCGTAATTCGATA >NR_080086 GATGAGCGGAGCACT >NR_070118 CGTAGCGAAACCGAG >NR_092384,>NR_098291 GGGGGGGAGGTCCGA >NR_021671,>NR_036907,>NR_080961 AGGGAGGGGGGTTGT >NR_026207 ATGGGGCAGACGCGA >NR_072314

    Benchmark - 8-core VM, CentOS 7.7

    62.883 seconds - op's code 33.594 seconds - search_db.pl 6.367 seconds - search_db_mce.pl 4.991 seconds - search_db_chunk.pl 8.093 seconds - search_db_inline_c.pl 1.530 seconds - search_db_inline_c_mce.pl

    Benchmark - 8-core VM, Xubuntu 18.04.3

    36.684 seconds - search_db.pl 7.403 seconds - search_db_mce.pl 5.986 seconds - search_db_chunk.pl 11.009 seconds - search_db_inline_c.pl 2.188 seconds - search_db_inline_c_mce.pl

    Benchmark - 8-core, macOS Mojave 10.14.6

    28.857 seconds - search_db.pl 6.130 seconds - search_db_mce.pl 5.219 seconds - search_db_chunk.pl 27.322 seconds - search_db_inline_c.pl 5.808 seconds - search_db_inline_c_mce.pl

    This completes the exercise. It boggles my mind comparing CentOS vs. Xubuntu. Ditto, Inline::C running faster on Linux vs. macOS. Testing was done with Perl 5.30.1. The virtualization is handled by VMware Fusion.

    Regards, Mario

Log In?
Username:
Password:

What's my password?
Create A New User
Node Status?
node history
Node Type: note [id://11110419]
help
Chatterbox?
and the web crawler heard nothing...

How do I use this? | Other CB clients
Other Users?
Others browsing the Monastery: (4)
As of 2020-09-27 01:03 GMT
Sections?
Information?
Find Nodes?
Leftovers?
    Voting Booth?
    If at first I donít succeed, I Ö










    Results (142 votes). Check out past polls.

    Notices?