parallel instances running issue

[Maltsev__Sergey]

Hi,

I've got a C++ OpenVino project run under Ubuntu 18.04.

It takes rtsp stream, decodes, makes some detection with default SSD architecture

Running just one instance of the project takes approx 30ms to process a frame.

2 instances - 80ms

3 - 150ms

4 - 210ms

5 - 280ms

Why it so non-linear running on a CPU only?

What is the proper way to run several instances on one CPU?

Thanks!

[Shubha_R_Intel]

Dear Sergey,

Have you looked at our async samples ? Please study our samples with _async in the name. Also please read the below blog-post (after downloading OpenVino 2019 R1). There are a few options you can try to improve OpenVino performance in alignment with physical CPU cores and thread instances.

https://software.intel.com/en-us/blogs/2019/04/02/improved-parallelization-extended-deep-learning-capabilities-in-intel-distribution

Thanks for using OpenVino !

Shubha