CP-SAT: Best machine type on GCP Compute Engine

[Rick Adriaensen]

Hi all,

I am working on a job shop scheduling problem, similar to the flexible job shop problem:

https://github.com/google/or-tools/blob/stable/examples/python/flexible_job_shop_sat.py

When increasing the number of jobs to be scheduled the computation time increases. For my problem it is desired to keep the computation time to a minimum. Therefore, I want to use a more powerfull machine on Google Cloud Platform (GCP) Compute Engine.

I can choose from the following machine types: https://gcpinstances.doit-intl.com/

How do I have to interpret the performance of the CP-SAT solver on these machines?

The more vCPUs, the better and/or the more memory the better? Or is there a threshold on the amount of vCPUs and memory the CP-SAT solver can use? 

Best regards,

Rick Adriaensen

[Kai Müller]

Also interested in that.

[Laurent Perron]

 I would recommend 16 true cores (not hyperthreaded), so most likely 32 CPU.

Memory is important too.

Perhaps c2-standard-30 or n2d-standard-32, t2d-standard-32

Laurent Perron | Operations Research | lpe...@google.com | (33) 1 42 68 53 00

--
You received this message because you are subscribed to the Google Groups "or-tools-discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to or-tools-discu...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/or-tools-discuss/b0d1a03c-ab99-4b05-a5bc-b5044397b9bdn%40googlegroups.com.

[Laurent Perron]

General ideas:

I am always speaking in true cores. So most likely take a machine with 2x cores vs num_search_workers.

minimum: 8 workers

16 workers: more LNS workers (faster good solution) and a few workers dedicated to improving the lower bound/proving optimality.

17 workers and more: only add LNS workers. Useful if finding very good solutions is really hard/slow.

As said before, we tune for 16 workers.

Laurent Perron | Operations Research | lpe...@google.com | (33) 1 42 68 53 00

--
You received this message because you are subscribed to the Google Groups "or-tools-discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to or-tools-discu...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/or-tools-discuss/5f8c7896-fdd2-4585-847a-3f9ddda80df9n%40googlegroups.com.