Mathematica 9 is now available
Services & Resources / Wolfram Forums / MathGroup Archive
-----

MathGroup Archive 2009

[Date Index] [Thread Index] [Author Index]

Search the Archive

Re: Parallel computing with Mathematica 7?

  • To: mathgroup at smc.vnet.net
  • Subject: [mg95387] Re: Parallel computing with Mathematica 7?
  • From: mark mcclure <mcmcclur at unca.edu>
  • Date: Sat, 17 Jan 2009 05:29:38 -0500 (EST)
  • References: <gkppvv$dsp$1@smc.vnet.net>

On Jan 16, 6:10 am, einsch... at gmail.com wrote:
> I have installed Mathematica 7 on my new laptop (Intel Centrino double
> core, Windows XP) and the first thing I tried was the long awaited
> parallllllel computing. I have tried different new commands and what I
> saw was a black box with input and no output. I could not see any
> evidence that parallelization works for me.

I'm not sure I quite understand your example, but here's an
operation that is sped up tremendously via parallelization:

t = AbsoluteTime[];
data = Table[PrimeQ[x + I*y, GaussianIntegers -> True],
   {x, -1000, 1000}, {y, -1000, 1000}];
time1 = AbsoluteTime[] - t
7.042566

To run it in parallel, it might make sense to LaunchKernels first:

LaunchKernels[]

I get 8 kernels on my machine.  Now, simply change the Table
to ParallelTable:

t = AbsoluteTime[];
data = ParallelTable[PrimeQ[x + I*y, GaussianIntegers -> True],
   {x, -1000, 1000}, {y, -1000, 1000}];
time2 = AbsoluteTime[] - t
1.238292

A silly example, perhaps, but there are quite a few similar
types of exercises at Project Euler:
http://projecteuler.net/

Also, I posted a webpage a month or so ago illustrating
a technique to use Mathematica to coordinate Java programs
through JLink:
http://facstaff.unca.edu/mcmcclur/Mathematica/ParallelJLink/


Finally, here are a couple more things to keep in mind with
regard to parallel computation:
 * Due to overhead in running parallel computations (extra
 memory, interprocess communication, etc.) it is very unusual
 for a parallel computation on n kernels to run n times faster.
 * There's generally no apriori way to break the problem up,
 since we don't necessarily know which sub-parts will run
 faster than other sub-parts.  It's best if, like in this
 example, the sub-problems run at similar rates.

Mark McClure


  • Prev by Date: Re: Which editor do you use for math articles
  • Next by Date: Re: Import MDB
  • Previous by thread: Re: Constrain independent variable in FindFit?
  • Next by thread: Re: Parallel computing with Mathematica 7?