Is there a way to start a parellel pool with maximum available workers as the cores on the machine?

Question

Leos Pohl 2021-8-24

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1439949-is-there-a-way-to-start-a-parellel-pool-with-maximum-available-workers-as-the-cores-on-the-machine

回答： Raymond Norris 2021-8-24

I am using a compiled code which then is being run on a cluster, each cluster node can have a different number of available cores. Is there a way to start parpool with the maximum possible amount of workers? If I start pool = parpool('local'); I only get 12 workers even though on the node, where such code is run, I have 32 cores allocated.

2 个评论
显示无隐藏无

KSSV 2021-8-24

Check the parallel preferences, there you can increase the number of workers.

Leos Pohl 2021-8-24

But this is compiled code.

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Raymond Norris 2021-8-24

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1439949-is-there-a-way-to-start-a-parellel-pool-with-maximum-available-workers-as-the-cores-on-the-machine#answer_773704

在 MATLAB Online 中打开

pool = parpool('local',maxNumCompThreads);

Alternatively, if you're running this through a scheduler (e.g. PBS), you could query the scheduler first.

% Query for available cores (assume either Slurm or PBS)                                                                                             
sz = str2num([getenv('SLURM_CPUS_PER_TASK') getenv('PBS_NP')]); %#ok<ST2NM>                                                                          
if isempty(sz)
    % Not running through a scheduler, get default size
    sz = maxNumCompThreads; end
pool = parpool('local',sz);