Beefy Boxes and Bandwidth Generously Provided by pair Networks vroom
The stupid question is the question not asked
 
PerlMonks  

Trimming benchmarks

by Tanktalus (Canon)
on Oct 14, 2010 at 14:42 UTC ( #865276=perlquestion: print w/ replies, xml ) Need Help??
Tanktalus has asked for the wisdom of the Perl Monks concerning the following question:

For some unknown reason, I was perusing old posts on StackOverflow, and came across an old benchmark I posted there. Since that time, I've upgraded CPU and installed multiple versions of perl, so I thought I'd see how much of an improvement I got. What I got was a shocker to me. Ok, that's a bit strong - it was still definitely a surprise.

My benchmark was originally tested on a Phenom 2.5GHz system with perl 5.8.8. I then tried on my i7 2.8GHz system with perl 5.12.2. Some of the codepaths improved, but some, like single, went way downhill. So I tried with 5.8.8 on the new box, too, and saw that this was a bigger impact.

I was hoping that regex speed would improve, or at least be statistically similar, from release to release. In some of these cases, however, that does not seem to be the case. I then went and tried all the versions I have and got some widely varying numbers. Is this what others would expect?

perl5.8.8 Rate single capture kramercap trick double single 3596/s -- -9% -9% -95% -96% capture 3946/s 10% -- -0% -94% -95% kramercap 3960/s 10% 0% -- -94% -95% trick 71359/s 1884% 1708% 1702% -- -11% double 80563/s 2140% 1942% 1935% 13% -- perl5.10.1 Rate single capture kramercap trick double single 1857/s -- -50% -51% -98% -98% capture 3702/s 99% -- -1% -95% -96% kramercap 3756/s 102% 1% -- -95% -96% trick 76323/s 4010% 1962% 1932% -- -10% double 85132/s 4485% 2200% 2167% 12% -- perl5.11.5 Rate single capture kramercap trick double single 1820/s -- -50% -51% -98% -98% capture 3657/s 101% -- -1% -95% -96% kramercap 3691/s 103% 1% -- -95% -96% trick 75045/s 4024% 1952% 1933% -- -11% double 84148/s 4524% 2201% 2180% 12% -- perl5.12.0 Rate single capture kramercap trick double single 1734/s -- -50% -50% -98% -98% capture 3446/s 99% -- -1% -95% -96% kramercap 3466/s 100% 1% -- -95% -96% trick 72542/s 4084% 2005% 1993% -- -10% double 80991/s 4572% 2250% 2237% 12% -- perl5.12.1 Rate single capture kramercap trick double single 1749/s -- -49% -51% -98% -98% capture 3455/s 98% -- -2% -95% -96% kramercap 3540/s 102% 2% -- -95% -96% trick 74052/s 4134% 2044% 1992% -- -10% double 82732/s 4630% 2295% 2237% 12% -- perl5.12.2 Rate single capture kramercap trick double single 1972/s -- -50% -51% -97% -97% capture 3915/s 99% -- -2% -94% -95% kramercap 3998/s 103% 2% -- -94% -95% trick 64740/s 3183% 1554% 1519% -- -11% double 73100/s 3607% 1767% 1728% 13% --
Since it's often deemed bad behaviour to link to code elsewhere (though I don't see StackOverflow being quite the same issue), here's a copy of the benchmark code.
#!/usr/bin/perl use strict; use warnings; use Benchmark qw(:all); my $a = 'a' x 1_000; my @x = ( " $a ", "$a ", $a, " $a" ); cmpthese(-5, { single => sub { for my $s (@x) { my $x = $s; $x =~ s/^\s+|\s+$//g; } }, double => sub { for my $s (@x) { my $x = $s; $x =~ s/^\s+//; $x =~ s/\s+$//; } }, trick => sub { for my $s (@x) { my $x = $s; s/^\s+//, s/\s+$// for $x; } }, capture => sub { for my $s (@x) { my $x = $s; $x =~ s/\A\s*(.*?)\s*\z/$1/ } }, kramercap => sub { for my $s (@x) { my $x = $s; ($x) = $x =~ /^\s*(.*?)\s*$/ } }, } );

Comment on Trimming benchmarks
Select or Download Code
Re: Trimming benchmarks
by moritz (Cardinal) on Oct 14, 2010 at 15:34 UTC
    One remark: the regex engine is fast, sub invocations are slow. To reduce the overhead of sub invocation, I'd replicate @x a few hundred or thousand times.
    I was hoping that regex speed would improve, or at least be statistically similar, from release to release. In some of these cases, however, that does not seem to be the case. I then went and tried all the versions I have and got some widely varying numbers. Is this what others would expect?

    As far as I remember, many changes to the regex engine improved its sanity with respect to Unicode strings - mostly their semantics, speed suffered a bit from time to time.

    So it's not too surprising that some of them got slower.

    In case of the "single" benchmark, it might be worth to clone the perl git repo, and use git bisect to find out which changeset it was that slowed it down that much - if it was a single one. Then tell p5p about your findings, maybe the performance loss can be tackled.

    Perl 6 - links to (nearly) everything that is Perl 6.

Log In?
Username:
Password:

What's my password?
Create A New User
Node Status?
node history
Node Type: perlquestion [id://865276]
Approved by tinita
help
Chatterbox?
and the web crawler heard nothing...

How do I use this? | Other CB clients
Other Users?
Others musing on the Monastery: (8)
As of 2014-04-17 05:46 GMT
Sections?
Information?
Find Nodes?
Leftovers?
    Voting Booth?

    April first is:







    Results (440 votes), past polls