Will parallel computing plus multiple matlab windows increase the processing efficiency?

Question

RUI Li 2023-9-26

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/2025772-will-parallel-computing-plus-multiple-matlab-windows-increase-the-processing-efficiency

评论： RUI Li 2023-9-27

I have a computer with 16 physical cores to compute a very complex task T. When I use the parallel pool (all 16 workers were used) to process the task, the CPU utilization is ~13%. If I want to increase the processing efficiency and the CPU utlization, is it possible to split the tasks (e.g., two halfs T1 and T2) for multiple Matlab windows in the same computer (e.g., allocate T1 and T2 for 2 Matlab windows, each of which also use the paralle pool with 16 workers)? Would it be faster than running in 1 matlab window?

Thanks in advance for any help!

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

Pratyush 2023-9-26

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/2025772-will-parallel-computing-plus-multiple-matlab-windows-increase-the-processing-efficiency#answer_1318762

Hi Rui,

Allocating different parts of a complex task to multiple instances of MATLAB running in separate windows on the same computer can potentially increase processing efficiency and CPU utilization.

By splitting the task T into two halves, T1 and T2, and allocating them to two MATLAB windows, each utilizing a parallel pool with 16 workers, you can effectively utilize all the available cores. This can lead to improved processing efficiency and higher CPU utilization compared to running the entire task in a single MATLAB window.

However, the overall speedup achieved by this approach depends on the nature of the task T and how well it can be divided into independent parts. If T can be easily partitioned into two halves that can be processed independently without significant communication or dependencies between them, then running them in separate MATLAB windows can provide a significant speedup.

Additionally, the speedup achieved by parallelism is subject to Amdahl's Law, which states that the maximum speedup is limited by the portion of the task that cannot be parallelized. If a significant portion of your task is inherently sequential and cannot be divided, the overall speedup may be less than expected.