Thread Subject: Parallel computing

Hi, there,

I am testing the performance of parfor and spmd on some simple codes using tic and toc. I found that the running time changed all the time. Could someone give me an idea on how to get a stable result and what is the difference between them. Thanks.

"terfil zhuang" <terfil@gmail.com> wrote in message <h5pjb9$cqj$1@fred.mathworks.com>...
> Hi, there,
>
> I am testing the performance of parfor and spmd on some simple codes using tic and toc. I found that the running time changed all the time. Could someone give me an idea on how to get a stable result and what is the difference between them. Thanks.

Terfil,

Have you looked at the product demos? They might be a good place to start, both for benchmarking in general and benchmarking parfor.

http://www.mathworks.com/products/parallel-computing/demos.html?file=/products/demos/shipping/distcomp/paralleldemo_parfor_bench.html
http://www.mathworks.com/products/parallel-computing/demos.html?file=/products/demos/shipping/distcomp/paralleldemo_distribjob_bench.html

Best,

Narfi

Also, the variability in your performance could have to do with you running other programs in the background. When I use the PCT, matlab will take up 100% of my CPU's resources unless other things are running in the backgorund, in which case it will take up whatever's left. Try running your code with nothing in the background.

"Juliette Salexa" <juliette.physicist@gmail.com> wrote in message <h603fc$oq7$1@fred.mathworks.com>...
> Also, the variability in your performance could have to do with you running other programs in the background. When I use the PCT, matlab will take up 100% of my CPU's resources unless other things are running in the backgorund, in which case it will take up whatever's left. Try running your code with nothing in the background.

Thanks for replying. After running several times, the results get stable. Also, the spmd does not work well as parfor. The code is showed as below:

clear;
clc;
tic
matlabpool open 2
A=zeros(10000000,1);
B=A;
spmd
   for i=1:size(A,1) %or using parfor
        B(i)=A(i)+i;
    end
end
matlabpool close
toc