parfor 内存累计不释放

Question

克 2023-7-22

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1999323-parfor

评论：克 2023-8-9

这样的parfor里面存在什么问题吗？

运行的服务器512G内存 10核运算仍然会满内存

n = 1; %as an input  is a content in parfor (常数)
ebler = 0;
for k = 1:loopout % loopout = 100000
    C = 0; V = 0;
    Blockerror = zeros(loopin,,1);
    parfor j = 1:loopin %loopin = 1000
        [C V] = function1();%renturn c and v  
        Blockerror(j,,1) = 1-qfunc(sqrt(n)*C./sqrt(V));
    end
    bler = bler + sum(Blockerror);%sum Blockerror  (累加Blockerror)
    clear Blockerror C V;
end

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

Shushant 2023-8-9

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1999323-parfor#answer_1285392

Hi 克,

I understand that when you execute your code your server memory is running out of space. This issue is because you are using a "parfor" loop inside a for loop, due to this the amount of data transferred will be much greater and you will also not notice any major computational benefit when compared to nested-for loops. This happens because each of the multiple “parfor” executions incurs an overhead.

To solve this issue run the outer loop in parallel instead of the inner loop, this will ensure that the overhead incurs only once. For more information on the same, kindly refer to the following documentation -

https://in.mathworks.com/help/parallel-computing/nested-parfor-loops-and-for-loops.html

I hope this provides you with the required information regarding your query.

Thank you,

Shushant

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

克 2023-8-9

Thanks for your answer. I find my problem and solved it by using outer loop "parfor" and changing the huge array to small array, then sum it. Thanks again!

请先登录，再进行评论。