From: Sam Roberts <sroberts@...>
Date: 2006-05-23T04:49:04+09:00
Subject: Re: Zed and Luis drop the bomb on Ruby's poor performance

Quoting james_b@neurogami.com, on Tue, May 23, 2006 at 01:50:34AM +0900:
> Those people having speed issues: Is it not a reasonable option to 
> replace the bottleneck Ruby code with custom compiled C libraries?

Its not obvious that this can always help, though it might in some cases.

I've been benchmarking vpim's icalendar decoding, and haven't found any
way to get it significantly faster yet.

Its a bit of an issue, a utility I use a lot to print out events and
todos for the next week usually takes 5-10 seconds to run. This is long
enough to notice!

When I tried running profile, and it told me 40% of the time was spent
in String case-insensitive comparisons, so I reworked the internals to
always uppercase strings so that wasn't necessary. Got about 5%.

I even tried using symbols instead of strings, I expected from lisp
experience that symbol comparisons would be really fast, but they
are only slightly faster (for strings of the size I'm working with
anyway). **

I ran out of time at that point, but its not clear to me what kind
of algorithmic changes I can make to speed things up, and I'm not too
sure where the time is being spent, other than in string operations.

There definitely isn't one part of the app I can point to and say
"recode that in C and my problems will go away". Also, you might have
noticed but ruby libraries that require extensions to be compiled tend
to not get a lot of traction in the ruby community.

Anyhow, I just spent a weekend on performance tuning, I might still find
out some clever way to restructure my code so that it gets twice as
fast, but a 2-3 times speed improvement would make a big difference to
my library. The difference between 6 seconds and under 2 seconds is huge
in human interface terms.

Cheers,
Sam


**

                          user     system      total        real
noop                  0.700000   0.000000   0.700000 (  0.698633)
String#==             2.410000   0.010000   2.420000 (  2.445288)
String#casecmp        2.650000   0.000000   2.650000 (  2.652620)
Symbol#==             2.280000   0.000000   2.280000 (  2.274505)
String#to_sym         2.460000   0.000000   2.460000 (  2.449362)
String#up             4.730000   0.000000   4.730000 (  4.721140)
String#up#to_sym      6.030000   0.000000   6.030000 (  5.986897)


-- bench.rb --
require 'benchmark'
require 'pp'

str0 = "a" * 20
str1 = str0.upcase
sym0 = str0.to_sym
sym1 = str1.to_sym

$N = 1000000

Benchmark.bm(20) do |x|

	x.report("noop")  do
    $N.times do
    end
  end

	x.report("String#==")  do
    $N.times do
      str0 == str1
    end
  end

	x.report("String#casecmp")  do
    $N.times do
      str0.casecmp(str1)
    end
  end

	x.report("Symbol#==")  do
    $N.times do
      sym0 == sym1
    end
  end

	x.report("String#to_sym")  do
    $N.times do
      str1.to_sym
    end
  end

	x.report("String#up")  do
    $N.times do
      str1.upcase
    end
  end

	x.report("String#up#to_sym")  do
    $N.times do
      str1.upcase.to_sym
    end
  end

end