http://www.perlmonks.org?node_id=658793


in reply to Re^2: 5:10: Why is this slower than 5.8.8?
in thread 5:10: Why is this slower than 5.8.8?

Well, then try the following script 
my $max = 50000;
my @l = ('a'..'z');
my @keys = map { join '', map $l[rand @l], 0..71 } 0..$max;

[download]
On my system running this script with a perlmalloc-built perl is 600x (!) faster than with a sysmalloc-built perl: 
$ time /usr/local/src/perl-5.10.0-perlmalloc/perl -I/usr/local/src/per
+l-5.10.0-perlmalloc/lib ~/trash/slow_malloc2.pl
3.027u 0.435s 0:03.68 93.7%     1448+324155k 0+0io 0pf+0w

$ time /usr/local/src/perl-5.10.0-sysmalloc/perl -I/usr/local/src/perl
+-5.10.0-sysmalloc/lib ~/trash/slow_malloc2.pl
405.090u 189.574s 13:01.52 76.0%        1466+495449k 0+0io 0pf+0w

[download]
This is a pathological case, but this may happen with FreeBSD's malloc.

Replies are listed 'Best First'.
Re^4: 5:10: Why is this slower than 5.8.8?
by Anonymous Monk on Dec 24, 2007 at 08:36 UTC
    You found one case on 10000. Other 9999 cases is slower with perl's malloc. And this is not the subject of the topic! Class::Accessor::Grouped anyway runs faster without perl's malloc and the problem is not in the type of malloc.
[reply]
      This does not seem to be the public opinion in this topic. For example, FreeBSD's perl port contains the lines: 
              @${ECHO} "      WITHOUT_PERL_MALLOC=yes   Use FreeBSD system m
+alloc"
        @${ECHO} "                                (uses less memory, b
+ut slower)

      [download]
      And the hint file in the perl distribution used to have the following lines: 
      4.*)   # In FreeBSD 4 and 5 the system malloc is performance-wise
       # VERY bad for Perl-- we are talking of differences of not
       # one, but TWO magnitudes.

      [download]
      This is not part of the default hint file anymore, but only because perl's malloc does not work with threaded builds, so Jarkko decided to always use the system's malloc.
[reply]
[d/l]
[select]

In Section Seekers of Perl Wisdom