[Haskell-cafe] Unbelievable parallel speedup

[Yves Parès]

> I've enjoyed reading Simon Marlow's new tutorial on parallel and
concurrent programming

I am interested: where I this tutorial?

2011/6/3 John D. Ramsdell <ramsdell0 at gmail.com>

> I've enjoyed reading Simon Marlow's new tutorial on parallel and
> concurrent programming, and learned some surprisingly basic tricks.  I
> didn't know about the '-s' runtime option for printing statistics.  I
> decided to compute speedups for a program I wrote just as Simon did,
> after running the program on an unloaded machine with four processors.
>  When I did, I found the speedup on two processors was 2.4, on three
> it was 3.2, and on four it was 4.4!  Am I living in a dream world?
>
> I ran the test nine more times, and here is a table of the speedups.
>
> 2.35975 3.42595 4.39351
> 1.57458 2.18623 2.94045
> 1.83232 2.77858 3.41629
> 1.58011 2.37084 2.94913
> 2.36678 3.63694 4.42066
> 1.58199 2.29053 2.95165
> 1.57656 2.34844 2.94683
> 1.58143 2.3242  2.95098
> 2.36703 3.36802 4.41918
> 1.58341 2.30123 2.93933
>
> That last line looks pretty reasonable to me, and is what I expected.
> Let's look at a table of the elapse times.
>
> 415.67  176.15  121.33  94.61
> 277.52  176.25  126.94  94.38
> 321.37  175.39  115.66  94.07
> 277.72  175.76  117.14  94.17
> 415.63  175.61  114.28  94.02
> 277.75  175.57  121.26  94.10
> 277.68  176.13  118.24  94.23
> 277.51  175.48  119.40  94.04
> 415.58  175.57  123.39  94.04
> 277.62  175.33  120.64  94.45
>
> Notice that the elapse times for two and four processors is pretty
> consistent, and the one for three processors is a little inconsistent,
> but the times for the single processor case are all over the map.  Can
> anyone explain all this variance?
>
> I have enclosed the raw output from the runs and the script that was
> run ten times to produce the output.
>
> John
>
> _______________________________________________
> Haskell-Cafe mailing list
> Haskell-Cafe at haskell.org
> http://www.haskell.org/mailman/listinfo/haskell-cafe
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.haskell.org/pipermail/haskell-cafe/attachments/20110603/5e713eed/attachment.htm>