I''d just gotten done implementing DEFLATE in pure ruby, and figuring there''d be a performance hit I was interested to see what it would be. On IronRuby (compiled in Release) it takes a full 16 seconds to decompress a ~1KB file... in CRuby it''s less then a second. An ~800KB file takes 42 seconds in CRuby, while IronRuby takes a whopping... well I don''t know, I killed it after it was running for about a half hour and using upwards of 100MB of memory! Now this is just a straight port of "puff.c" from the zlib library, so it''s certainly not optimized for speed in any way, but the difference in runtimes really struck me. I suppose at the very least we have a good stress test for rbx ;) Just thought I''d share. -- Michael Letterle [Polymath Prokrammer] http://blog.prokrams.com
Michael Letterle:> I''d just gotten done implementing DEFLATE in pure ruby, and figuring > there''d be a performance hit I was interested to see what it would be. > On IronRuby (compiled in Release) it takes a full 16 seconds to > decompress a ~1KB file... in CRuby it''s less then a second. An ~800KB > file takes 42 seconds in CRuby, while IronRuby takes a whopping... > well I don''t know, I killed it after it was running for about a half > hour and using upwards of 100MB of memory! > > Now this is just a straight port of "puff.c" from the zlib library, so > it''s certainly not optimized for speed in any way, but the difference > in runtimes really struck me. > > I suppose at the very least we have a good stress test for rbx ;) Just > thought I''d share.Cool :) Can you send along the source code for your implementation so we can investigate? Thanks -John
I''ll be posting it up (probably on RubyForge) once I get cleaned up a bit (acknowledgments, rdoc stuff), if you want the raw stuff I can email it straight to you. On Mon, Mar 10, 2008 at 5:33 PM, John Lam (DLR) <jflam at microsoft.com> wrote:> Michael Letterle: > > > > > I''d just gotten done implementing DEFLATE in pure ruby, and figuring > > there''d be a performance hit I was interested to see what it would be. > > On IronRuby (compiled in Release) it takes a full 16 seconds to > > decompress a ~1KB file... in CRuby it''s less then a second. An ~800KB > > file takes 42 seconds in CRuby, while IronRuby takes a whopping... > > well I don''t know, I killed it after it was running for about a half > > hour and using upwards of 100MB of memory! > > > > Now this is just a straight port of "puff.c" from the zlib library, so > > it''s certainly not optimized for speed in any way, but the difference > > in runtimes really struck me. > > > > I suppose at the very least we have a good stress test for rbx ;) Just > > thought I''d share. > > Cool :) Can you send along the source code for your implementation so we can investigate? > > Thanks > -John > > _______________________________________________ > Ironruby-core mailing list > Ironruby-core at rubyforge.org > http://rubyforge.org/mailman/listinfo/ironruby-core >-- Michael Letterle [Polymath Prokrammer] http://blog.prokrams.com
I bet you a nickel that if you replace a few key block loops (i.e. each, times, etc) with a vanilla while loop that you''ll see a 10x speedup... I''ve seen some really huge performance hits on IR using blocks. -Eric On Mon, Mar 10, 2008 at 6:06 PM, Michael Letterle < michael.letterle at gmail.com> wrote:> I''ll be posting it up (probably on RubyForge) once I get cleaned up a > bit (acknowledgments, rdoc stuff), if you want the raw stuff I can > email it straight to you. > > On Mon, Mar 10, 2008 at 5:33 PM, John Lam (DLR) <jflam at microsoft.com> > wrote: > > Michael Letterle: > > > > > > > > > I''d just gotten done implementing DEFLATE in pure ruby, and figuring > > > there''d be a performance hit I was interested to see what it would > be. > > > On IronRuby (compiled in Release) it takes a full 16 seconds to > > > decompress a ~1KB file... in CRuby it''s less then a second. An > ~800KB > > > file takes 42 seconds in CRuby, while IronRuby takes a whopping... > > > well I don''t know, I killed it after it was running for about a half > > > hour and using upwards of 100MB of memory! > > > > > > Now this is just a straight port of "puff.c" from the zlib library, > so > > > it''s certainly not optimized for speed in any way, but the difference > > > in runtimes really struck me. > > > > > > I suppose at the very least we have a good stress test for rbx ;) > Just > > > thought I''d share. > > > > Cool :) Can you send along the source code for your implementation so > we can investigate? > > > > Thanks > > -John > > > > _______________________________________________ > > Ironruby-core mailing list > > Ironruby-core at rubyforge.org > > http://rubyforge.org/mailman/listinfo/ironruby-core > > > > > > -- > Michael Letterle > [Polymath Prokrammer] > http://blog.prokrams.com > _______________________________________________ > Ironruby-core mailing list > Ironruby-core at rubyforge.org > http://rubyforge.org/mailman/listinfo/ironruby-core >-------------- next part -------------- An HTML attachment was scrubbed... URL: http://rubyforge.org/pipermail/ironruby-core/attachments/20080310/98f593a6/attachment.html
Don''t know what I''m going to do with a nickel... Actually since it''s basically a port of a c program I''m only really using blocks in two places, input and output, plus on other little place.. I can try but I''d be surprised.. unless for loops in a range count... On Mon, Mar 10, 2008 at 6:37 PM, Eric Nicholson <enicholson at gmail.com> wrote:> I bet you a nickel that if you replace a few key block loops (i.e. each, > times, etc) with a vanilla while loop that you''ll see a 10x speedup... I''ve > seen some really huge performance hits on IR using blocks. > > -Eric > > > On Mon, Mar 10, 2008 at 6:06 PM, Michael Letterle > <michael.letterle at gmail.com> wrote: > > I''ll be posting it up (probably on RubyForge) once I get cleaned up a > > bit (acknowledgments, rdoc stuff), if you want the raw stuff I can > > email it straight to you. > > > > > > > > > > On Mon, Mar 10, 2008 at 5:33 PM, John Lam (DLR) <jflam at microsoft.com> > wrote: > > > Michael Letterle: > > > > > > > > > > > > > I''d just gotten done implementing DEFLATE in pure ruby, and figuring > > > > there''d be a performance hit I was interested to see what it would > be. > > > > On IronRuby (compiled in Release) it takes a full 16 seconds to > > > > decompress a ~1KB file... in CRuby it''s less then a second. An > ~800KB > > > > file takes 42 seconds in CRuby, while IronRuby takes a whopping... > > > > well I don''t know, I killed it after it was running for about a half > > > > hour and using upwards of 100MB of memory! > > > > > > > > Now this is just a straight port of "puff.c" from the zlib library, > so > > > > it''s certainly not optimized for speed in any way, but the difference > > > > in runtimes really struck me. > > > > > > > > I suppose at the very least we have a good stress test for rbx ;) > Just > > > > thought I''d share. > > > > > > Cool :) Can you send along the source code for your implementation so > we can investigate? > > > > > > Thanks > > > -John > > > > > > _______________________________________________ > > > Ironruby-core mailing list > > > Ironruby-core at rubyforge.org > > > http://rubyforge.org/mailman/listinfo/ironruby-core > > > > > > > > > > > > > -- > > Michael Letterle > > [Polymath Prokrammer] > > http://blog.prokrams.com > > _______________________________________________ > > > > > > > > Ironruby-core mailing list > > Ironruby-core at rubyforge.org > > http://rubyforge.org/mailman/listinfo/ironruby-core > > > > > _______________________________________________ > Ironruby-core mailing list > Ironruby-core at rubyforge.org > http://rubyforge.org/mailman/listinfo/ironruby-core > >-- Michael Letterle [Polymath Prokrammer] http://blog.prokrams.com
Alright, I''ll bet you a beer next time you''re in Raleigh, NC. :) For... in loops seem to use blocks. Anyway, in a quick test they seem to be about 200 - 1000 times slower than an equivalent while loop. A quick test that runs through 1000 iterations 2000 times takes 0.3s with a while loop, and 9.0s with a for loop on my system. Try a while loop... -Eric On Mon, Mar 10, 2008 at 6:44 PM, Michael Letterle < michael.letterle at gmail.com> wrote:> Don''t know what I''m going to do with a nickel... > > Actually since it''s basically a port of a c program I''m only really > using blocks in two places, input and output, plus on other little > place.. I can try but I''d be surprised.. unless for loops in a range > count... > > On Mon, Mar 10, 2008 at 6:37 PM, Eric Nicholson <enicholson at gmail.com> > wrote: > > I bet you a nickel that if you replace a few key block loops (i.e. each, > > times, etc) with a vanilla while loop that you''ll see a 10x speedup... > I''ve > > seen some really huge performance hits on IR using blocks. > > > > -Eric > > > > > > On Mon, Mar 10, 2008 at 6:06 PM, Michael Letterle > > <michael.letterle at gmail.com> wrote: > > > I''ll be posting it up (probably on RubyForge) once I get cleaned up a > > > bit (acknowledgments, rdoc stuff), if you want the raw stuff I can > > > email it straight to you. > > > > > > > > > > > > > > > On Mon, Mar 10, 2008 at 5:33 PM, John Lam (DLR) <jflam at microsoft.com> > > wrote: > > > > Michael Letterle: > > > > > > > > > > > > > > > > > I''d just gotten done implementing DEFLATE in pure ruby, and > figuring > > > > > there''d be a performance hit I was interested to see what it > would > > be. > > > > > On IronRuby (compiled in Release) it takes a full 16 seconds to > > > > > decompress a ~1KB file... in CRuby it''s less then a second. An > > ~800KB > > > > > file takes 42 seconds in CRuby, while IronRuby takes a > whopping... > > > > > well I don''t know, I killed it after it was running for about a > half > > > > > hour and using upwards of 100MB of memory! > > > > > > > > > > Now this is just a straight port of "puff.c" from the zlib > library, > > so > > > > > it''s certainly not optimized for speed in any way, but the > difference > > > > > in runtimes really struck me. > > > > > > > > > > I suppose at the very least we have a good stress test for rbx ;) > > Just > > > > > thought I''d share. > > > > > > > > Cool :) Can you send along the source code for your implementation > so > > we can investigate? > > > > > > > > Thanks > > > > -John > > > > > > > > _______________________________________________ > > > > Ironruby-core mailing list > > > > Ironruby-core at rubyforge.org > > > > http://rubyforge.org/mailman/listinfo/ironruby-core > > > > > > > > > > > > > > > > > > > -- > > > Michael Letterle > > > [Polymath Prokrammer] > > > http://blog.prokrams.com > > > _______________________________________________ > > > > > > > > > > > > Ironruby-core mailing list > > > Ironruby-core at rubyforge.org > > > http://rubyforge.org/mailman/listinfo/ironruby-core > > > > > > > > > _______________________________________________ > > Ironruby-core mailing list > > Ironruby-core at rubyforge.org > > http://rubyforge.org/mailman/listinfo/ironruby-core > > > > > > > > -- > Michael Letterle > [Polymath Prokrammer] > http://blog.prokrams.com > _______________________________________________ > Ironruby-core mailing list > Ironruby-core at rubyforge.org > http://rubyforge.org/mailman/listinfo/ironruby-core >-------------- next part -------------- An HTML attachment was scrubbed... URL: http://rubyforge.org/pipermail/ironruby-core/attachments/20080311/e2cfc2d3/attachment.html
So I did some testing on this, I wrapped all loops in an outer while
loop and ran them though both rbx and cruby:
i.e.
t = 0
while t < 2000
for i in (1..1000)
end
t+=1
end
cruby:
while: 0:01.22
for: 0:00.46
times: 0:00.75
rbx:
while: 0:01.83
for: 0:24.53
times: 0:24.84
Of course this is just a couple of runs, it would be more accurate to
do a whole suite and average them out, but I ran them a few times and
the times were relatively consistent.
Work to be done...
On Tue, Mar 11, 2008 at 9:57 AM, Eric Nicholson <enicholson at gmail.com>
wrote:> Alright, I''ll bet you a beer next time you''re in Raleigh,
NC. :)
>
> For... in loops seem to use blocks. Anyway, in a quick test they seem to
be
> about 200 - 1000 times slower than an equivalent while loop. A quick test
> that runs through 1000 iterations 2000 times takes 0.3s with a while loop,
> and 9.0s with a for loop on my system.
>
> Try a while loop...
> -Eric
>
>
>
> On Mon, Mar 10, 2008 at 6:44 PM, Michael Letterle
> <michael.letterle at gmail.com> wrote:
> > Don''t know what I''m going to do with a nickel...
> >
> > Actually since it''s basically a port of a c program
I''m only really
> > using blocks in two places, input and output, plus on other little
> > place.. I can try but I''d be surprised.. unless for loops in
a range
> > count...
> >
> >
> >
> >
> > On Mon, Mar 10, 2008 at 6:37 PM, Eric Nicholson <enicholson at
gmail.com>
> wrote:
> > > I bet you a nickel that if you replace a few key block loops
(i.e. each,
> > > times, etc) with a vanilla while loop that you''ll see a
10x speedup...
> I''ve
> > > seen some really huge performance hits on IR using blocks.
> > >
> > > -Eric
> > >
> > >
> > > On Mon, Mar 10, 2008 at 6:06 PM, Michael Letterle
> > > <michael.letterle at gmail.com> wrote:
> > > > I''ll be posting it up (probably on RubyForge) once
I get cleaned up a
> > > > bit (acknowledgments, rdoc stuff), if you want the raw stuff
I can
> > > > email it straight to you.
> > > >
> > > >
> > > >
> > > >
> > > > On Mon, Mar 10, 2008 at 5:33 PM, John Lam (DLR) <jflam at
microsoft.com>
> > > wrote:
> > > > > Michael Letterle:
> > > > >
> > > > >
> > > > >
> > > > > > I''d just gotten done implementing
DEFLATE in pure ruby, and
> figuring
> > > > > > there''d be a performance hit I was
interested to see what it
> would
> > > be.
> > > > > > On IronRuby (compiled in Release) it takes a
full 16 seconds to
> > > > > > decompress a ~1KB file... in CRuby it''s
less then a second. An
> > > ~800KB
> > > > > > file takes 42 seconds in CRuby, while IronRuby
takes a
> whopping...
> > > > > > well I don''t know, I killed it after it
was running for about a
> half
> > > > > > hour and using upwards of 100MB of memory!
> > > > > >
> > > > > > Now this is just a straight port of
"puff.c" from the zlib
> library,
> > > so
> > > > > > it''s certainly not optimized for speed
in any way, but the
> difference
> > > > > > in runtimes really struck me.
> > > > > >
> > > > > > I suppose at the very least we have a good stress
test for rbx ;)
> > > Just
> > > > > > thought I''d share.
> > > > >
> > > > > Cool :) Can you send along the source code for your
implementation
> so
> > > we can investigate?
> > > > >
> > > > > Thanks
> > > > > -John
> > > > >
> > > > > _______________________________________________
> > > > > Ironruby-core mailing list
> > > > > Ironruby-core at rubyforge.org
> > > > > http://rubyforge.org/mailman/listinfo/ironruby-core
> > > > >
> > > >
> > > >
> > > >
> > > >
> > > > --
> > > > Michael Letterle
> > > > [Polymath Prokrammer]
> > > > http://blog.prokrams.com
> > > > _______________________________________________
> > > >
> > > >
> > > >
> > > > Ironruby-core mailing list
> > > > Ironruby-core at rubyforge.org
> > > > http://rubyforge.org/mailman/listinfo/ironruby-core
> > > >
> > >
> > >
> > > _______________________________________________
> > > Ironruby-core mailing list
> > > Ironruby-core at rubyforge.org
> > > http://rubyforge.org/mailman/listinfo/ironruby-core
> > >
> > >
> >
> >
> >
> > --
> > Michael Letterle
> > [Polymath Prokrammer]
> > http://blog.prokrams.com
> > _______________________________________________
> > Ironruby-core mailing list
> > Ironruby-core at rubyforge.org
> > http://rubyforge.org/mailman/listinfo/ironruby-core
> >
>
>
> _______________________________________________
> Ironruby-core mailing list
> Ironruby-core at rubyforge.org
> http://rubyforge.org/mailman/listinfo/ironruby-core
>
>
--
Michael Letterle
[Polymath Prokrammer]
http://blog.prokrams.com