parfor works on local computer but not on a cluster

1 次查看（过去 30 天）

Christopher 2015-11-5

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/253157-parfor-works-on-local-computer-but-not-on-a-cluster

评论： Christopher 2015-11-5

My code has the following form:

 clear all
 rng('shuffle')
 runtables = importdata('runtable.mat'); % a 40000x22 table of 22 parameters for 40000 different models to run in parallel (embarassingly simple type)
 numfiles  = numel(runtables(:,1));
 runind    = 3000:numfiles; % run models 3000:40000
 % function importing data that is the same for all models (prevents importing of 500 MB of data thousands of times)
 [O,P,normx,normy,normz,refractmat,away_area] = importraddata3Dphaseadhoc(5,mindist_thresh,[1 1 1 1 1],[1 1 1 1 1]); 
 nummods    = numel(runind); % total number of models to run
 runtables2 = runtables(runind,:); % table of parameters only for models to run
 % BEGIN CYCLING THROUGH MODELS
 parfor j=1:nummods
 curmodel = runind(j);
    % import properties of current case model
    param1    = runtables2(j,1);     % parameter 1
    param2    = runtables2(j,2);     % parameter 2
    % etc.
 % begin numerical analysis for current case model.
 parsave(savefilename,Esum,extdisti,extdist,Estp,freq_min,freq_max,radintensity,polys,effk); % function to save case model results to HD
 end

My problem is that the code works perfectly fine when run on a local machine with 16 cores. However, having tried to run the code on a cluster, I have been unsuccessful. The problem is that when it gets to the parfor line (parfor j=1:nummods) it returns the error:

Index exceeds matrix dimensions

I have no idea why this occurs. On the other hand, the code runs correctly if I modify the code such that nummods==1.

The following link seems to address a problem like the one I am experiencing:

 http://www.mathworks.com/matlabcentral/answers/224150-index-exceeds-matrix-dimensions-error-when-using-parfor

but, I do not understand what to do, and I do not understand why my code works perfectly on a single multi-core machine but fails on a cluster.

Thanks.

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

Christopher 2015-11-5

在 MATLAB Online 中打开

OK, it works on the cluster if I change the initial lines to:

 clear all
 rng('shuffle')
 runtables = importdata('runtable9.mat');
 [O,P,normx,normy,normz,refractmat,away_area]   = importraddata3Dphaseadhoc(5,mindist_thresh,[1 1 1 1 1],[1 1 1 1 1]); % 
 nummods = numel(runtables(:,1));
 parfor j=1:nummods
 % etc.