Comments on: Playing with GHC’s parallel runtime http://donsbot.wordpress.com/2009/03/04/playing-with-ghcs-parallel-runtime/ A Journal of Haskell Programming Mon, 03 Oct 2011 02:09:43 +0000 hourly 1 http://wordpress.com/ By: Boids Simulation: Part 5 « Communicating Haskell Processes http://donsbot.wordpress.com/2009/03/04/playing-with-ghcs-parallel-runtime/#comment-190 Wed, 16 Sep 2009 15:47:34 +0000 http://donsbot.wordpress.com/?p=151#comment-190 [...] If your performance numbers are very odd, look to the garbage collector. (Don Stewart also recommends keeping an eye on GC when trying to get parallel speed-up.) [...]

]]> By: Isaac Gouy http://donsbot.wordpress.com/2009/03/04/playing-with-ghcs-parallel-runtime/#comment-58 Thu, 05 Mar 2009 18:06:04 +0000 http://donsbot.wordpress.com/?p=151#comment-58 After 5 years I still haven’t figured out how to make it clear how much data and analysis is there ;-(

I’ll update the u64 measurements today.

]]> By: dons00 http://donsbot.wordpress.com/2009/03/04/playing-with-ghcs-parallel-runtime/#comment-57 Thu, 05 Mar 2009 16:48:51 +0000 http://donsbot.wordpress.com/?p=151#comment-57 Interesting @Igouy! I’d not seen that table.

]]> By: Isaac Gouy http://donsbot.wordpress.com/2009/03/04/playing-with-ghcs-parallel-runtime/#comment-56 Thu, 05 Mar 2009 16:44:29 +0000 http://donsbot.wordpress.com/?p=151#comment-56 @ dons Meanwhile, on the single core…

Your “single core” URLS are the old Pentium 4 measurements – why not the forced-onto-one-core measurements from the same machine as the quad-core measurements?

http://shootout.alioth.debian.org/u64/benchmark.php?test=binarytrees&lang=all

]]> By: Simon Marlow http://donsbot.wordpress.com/2009/03/04/playing-with-ghcs-parallel-runtime/#comment-53 Wed, 04 Mar 2009 19:38:21 +0000 http://donsbot.wordpress.com/?p=151#comment-53 I should point out that some of the improvements mentioned in the “Runtime Support for Multicore Haskell” haven’t hit the HEAD yet, so you’re not seeing some of the benefit from the parallel GC. In particular what you have is similar to PARGC1 (terminology from the paper). Although on the binary trees program it doesn’t make a huge difference, the biggest gains here are to be had by just using a bigger heap, and not copying all those trees around.

After 6.10.2 is out I’ll focus on testing and cleaning up those patches and get them in.

]]> By: dons00 http://donsbot.wordpress.com/2009/03/04/playing-with-ghcs-parallel-runtime/#comment-52 Wed, 04 Mar 2009 18:42:32 +0000 http://donsbot.wordpress.com/?p=151#comment-52 @Sebastian

`using` is nicer syntax. I should have done that.

]]> By: Sebastian http://donsbot.wordpress.com/2009/03/04/playing-with-ghcs-parallel-runtime/#comment-51 Wed, 04 Mar 2009 18:39:17 +0000 http://donsbot.wordpress.com/?p=151#comment-51 Why do you do

parMap rnf id $ depth minN maxN

and not

depth minN maxN `using` parList rnf

?

]]> By: Mattias http://donsbot.wordpress.com/2009/03/04/playing-with-ghcs-parallel-runtime/#comment-50 Wed, 04 Mar 2009 17:40:55 +0000 http://donsbot.wordpress.com/?p=151#comment-50 Apparently one of the o’caml-entries¹ was rejected for setting a custom heap size. Interesting read none the less!

1: http://shootout.alioth.debian.org/gp4/benchmark.php?test=binarytrees&lang=ocaml&id=4

]]>