Overhead in parallel computing toolbox

8 次查看（过去 30 天）
Gonzalo Blanco
Gonzalo Blanco 2016-6-2
回答： Walter Roberson 2016-6-2
Hi, I have a problem with a parallel process.
I am calling a simulink model with a parfor loop to perform a kind of a montecarlo simulation.
if I run a scaling test, I obtain the following results in terms of speedup:
The hardware I'm using has two processors (Intel® Xeon® Processor E5-2630 v3). it means 8+8 physical cores.
It seems that there is any kind of overhead when using the two processors in the same parallel pool.
am I right? if yes, is there any way to solve or workaround this issue?

请先登录，再回答此问题。

回答（1 个）

Walter Roberson
Walter Roberson 2016-6-2
Yes, there can be a fair bit of overhead in parallel workers.
What you would ideally prefer to do is transfer as little data as possible with the workers, and have them do as much work as possible: you do this so that you amortize the overhead.
What you really do not want to do is transfer a lot of data and have the worker do very little: that wastes all of the time on setting up the task.

类别

在 Help Center 和 File Exchange 中查找有关 Multicore Processor Targets 的更多信息

标签

产品

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by