From: Charles Oliver Nutter <headius@...>
Date: 2010-07-22T14:39:23+09:00
Subject: Re: Count occurences of vars in array

On Tue, Jul 20, 2010 at 8:32 AM, Colin Bartlett <colinb2r@googlemail.com> wrote:
> jruby 1.5.0.RC1 (ruby 1.8.7 patchlevel 249) (2010-04-14 0b08bc7)
>  (Java HotSpot(TM) Client VM 1.6.0_14) [x86-java]
> owtdi      1.980000   0.000000   1.980000 (  1.979000)
> tmtowtdi   1.523000   0.000000   1.523000 (  1.522000)
> I ought to point out that in a subsequent JRuby run
>  owtdi user = 1.625 and tmtowtdi user = 1.809,
>  and in another run owtdi user = 2.118 and tmtowtdi user = 1.987,
> so quite variable! But there was quite a lot of apparently random and
> purposeless disk i/o going on while the benchmarks were running. (Thank you,
> Windows Vista!) So I would place even less reliance than usual on these
> benchmarks. But using inject does seem significantly slower, especially so
> using 1.8.6.

I know this isn't what the original post was about...but...

The variability is probably due partially to the jitting behavior of
the JVM; it's hard to know when the JVM will hit its stable state.

FWIW, a longer run will get even faster, and passing --server to JRuby
boosts it too:

jruby Java 6 server, fifth iteration:
                user     system      total        real
owtdi       0.223000   0.000000   0.223000 (  0.223000)
tmtowtdi    0.223000   0.000000   0.223000 (  0.223000)

ruby 1.9.2 trunkish:
                user     system      total        real
owtdi       0.470000   0.010000   0.480000 (  0.481433)
tmtowtdi    0.350000   0.000000   0.350000 (  0.350423)

I don't have an explanation for the performance difference; the logic
involved should be pretty native-lib heavy.

- Charlie