The parallel GC currently doesn't behave well with concurrent programs that uses multiple capabilities (aka OS threads), and the behaviour you see is the known symptom of this.. I believe that Simon Marlow has some fixes in hand that may go into 6.12.2.<div>
<br></div><div>Are you saying that you see two different classes of undesirable performance, one with -qg and one without? How are your threads in your real program communicating with each other? We've seen problems there when there's a lot of contention for e.g. IORefs among thousands of threads.<br>
<br><div class="gmail_quote">On Mon, Mar 1, 2010 at 7:59 AM, Michael Lesniak <span dir="ltr"><<a href="mailto:mlesniak@uni-kassel.de">mlesniak@uni-kassel.de</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
Hello haskell-cafe,<br>
<br>
Sorry for this long post, but I can't think of a way to describe and explain<br>
the problem in a shorter way.<br>
<br>
I've (again) a very strange behaviour with the parallel GC and would be glad<br>
if someone could either reproduce (and explain) it or provide a solution. A<br>
similar but unrelated problem has been described in [1].<br>
<br>
<br>
EXAMPLE CODE<br>
The following demonstration program, which is a much smaller and<br>
single-threaded version of my real problem behaves as my real program.<br>
It does some number crunching by calculating pi to a definable precision:<br>
<br>
> -- File Pi.hs<br>
> -- you need the numbers package from hackage.<br>
> module Main where<br>
> import Data.Number.CReal<br>
> import System.Environment<br>
> import GHC.Conc<br>
><br>
> main = do<br>
> digits <- (read . head) `fmap` getArgs :: IO Int<br>
> calcPi digits<br>
><br>
> calcPi digits = showCReal (fromEnum digits) pi `pseq` return ()<br>
<br>
Compile it with<br>
<br>
ghc --make -threaded -O2 Pi.hs -o pi<br>
<br>
<br>
BENCHMARKS<br>
On my two-core machine I get the following quite strange and<br>
unpredictable results:<br>
<br>
* Using one thread:<br>
<br>
$ for i in `seq 1 5`;do time pi 5000 +RTS -N1;done<br>
<br>
real 0m1.441s<br>
user 0m1.390s<br>
sys 0m0.020s<br>
<br>
real 0m1.449s<br>
user 0m1.390s<br>
sys 0m0.000s<br>
<br>
real 0m1.399s<br>
user 0m1.370s<br>
sys 0m0.010s<br>
<br>
real 0m1.401s<br>
user 0m1.380s<br>
sys 0m0.000s<br>
<br>
real 0m1.404s<br>
user 0m1.380s<br>
sys 0m0.000s<br>
<br>
<br>
* Using two threads, hence the parallel GC is used:<br>
<br>
for i in `seq 1 5`;do time pi 5000 +RTS -N2;done<br>
<br>
real 0m2.540s<br>
user 0m2.490s<br>
sys 0m0.010s<br>
<br>
real 0m1.527s<br>
user 0m1.530s<br>
sys 0m0.010s<br>
<br>
real 0m1.966s<br>
user 0m1.900s<br>
sys 0m0.010s<br>
<br>
real 0m5.670s<br>
user 0m5.620s<br>
sys 0m0.010s<br>
<br>
real 0m2.966s<br>
user 0m2.910s<br>
sys 0m0.020s<br>
<br>
<br>
* Using two threads, but disabling the parallel GC:<br>
<br>
for i in `seq 1 5`;do time pi 5000 +RTS -N2 -qg;done<br>
<br>
real 0m1.383s<br>
user 0m1.380s<br>
sys 0m0.010s<br>
<br>
real 0m1.420s<br>
user 0m1.360s<br>
sys 0m0.010s<br>
<br>
real 0m1.406s<br>
user 0m1.360s<br>
sys 0m0.010s<br>
<br>
real 0m1.421s<br>
user 0m1.380s<br>
sys 0m0.000s<br>
<br>
real 0m1.360s<br>
user 0m1.360s<br>
sys 0m0.000s<br>
<br>
<br>
THREADSCOPE<br>
I've additionally attached the threadscope profile of a really bad run,<br>
started with<br>
<br>
$ time pi 5000 +RTS -N2 -ls<br>
<br>
real 0m15.594s<br>
user 0m15.490s<br>
sys 0m0.010s<br>
<br>
as file pi.pdf<br>
<br>
<br>
FURTHER INFORMATION/QUESTION<br>
Just disabling the parallel GC leads to very bad performance in my original<br>
code, which forks threads with forkIO and does a lot of communications. Hence,<br>
using -qg is not a real option for me.<br>
<br>
Do I have overlooked some cruical aspect of this problem? If you've<br>
read this far, thank you for reading ... this far ;-)<br>
<br>
Cheers,<br>
Michael<br>
<br>
<br>
<br>
[1] <a href="http://osdir.com/ml/haskell-cafe@haskell.org/2010-02/msg00850.html" target="_blank">http://osdir.com/ml/haskell-cafe@haskell.org/2010-02/msg00850.html</a><br>
<br>
<br>
--<br>
Dipl.-Inf. Michael C. Lesniak<br>
University of Kassel<br>
Programming Languages / Methodologies Research Group<br>
Department of Computer Science and Electrical Engineering<br>
<br>
Wilhelmshöher Allee 73<br>
34121 Kassel<br>
<br>
Phone: +49-(0)561-804-6269<br>
<br>_______________________________________________<br>
Haskell-Cafe mailing list<br>
<a href="mailto:Haskell-Cafe@haskell.org">Haskell-Cafe@haskell.org</a><br>
<a href="http://www.haskell.org/mailman/listinfo/haskell-cafe" target="_blank">http://www.haskell.org/mailman/listinfo/haskell-cafe</a><br>
<br></blockquote></div><br></div>