Batch sounds slower than interactive parallel jobs

Question

Sylvain Rousseau 2013-7-9

1
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/81562-batch-sounds-slower-than-interactive-parallel-jobs

采纳的回答： Sean de Wolski

在 MATLAB Online 中打开

Hi !

I use Matlab 7.10.0 and Parallel Toolbox on a 8 cores blade.

Issuing that command :

tic;job=batch('birthdayscript','matlabpool',4);wait(job);ovrHead=toc;

I obtain ovrHead = 10.42 seconds

Issuing that command :

matlabpool open 4;tic;p = prunbirthday(2e4, 30), ovrHead=toc;matlabpool close

I obtain ovrHead = 1.39 seconds

May I ask you why batch method is so ineffective ?

Sylvain

PS : here the code of birthdayscript :

numtrials = 2e4;
groupsize = 30;
% Preallocate some memory for the matches
matches = zeros(1, numtrials);
parfor trial = 1:numtrials
    % Run a simulation for a group
    matches(trial) = birthday(groupsize);   
end
% Probability is the sum of matches divided by number of trials
prob = sum(matches)/numtrials;

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Sean de Wolski 2013-7-9

1
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/81562-batch-sounds-slower-than-interactive-parallel-jobs#answer_91273

编辑：Sean de Wolski 2013-7-9

batch is opening its own MATLAB pool of four workers so this is included in the timing of the first snippet and not the second.

2 个评论
显示无隐藏无

Sylvain Rousseau 2013-7-10

Many thanks Sean, it's crystal clear now !

Sean de Wolski 2013-7-10

Glad to hear. Good question too!

请先登录，再进行评论。

Answer 2

Sylvain Rousseau 2013-7-10

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/81562-batch-sounds-slower-than-interactive-parallel-jobs#answer_91334

编辑：Sylvain Rousseau 2013-7-10

在 MATLAB Online 中打开

Dear Sean,

Thanks for your answer.

That's what I thought at first sight, but the task overhead is the same when I move the tic after the batch() statement :

job=batch('birthdayscript','matlabpool',4);tic;wait(job);ovrHead=toc;

Here again, I obtain ovrHead = 10.42 seconds

Moreover, if I increase workers number up to seven (findResource gives a clusterSize of 8) , the overhead stays the same although I'am the only user of the computer.

Cheers.

sylvain