Originally posted by: AndyHam

Dear IBM,

Thanks for this great modeling tool. I have been applying CP to many combinatorial problems successfully owing to this forum. Recently, I have been working on Amazon/Walmart Flying Warehouse Scheduling problem, surrounded by large airships (airborne warehouse), small airships (shuttle), and drones (delivery). The great expressibility of CP made me successfully model this problem too. Now, I am trying to get the optimal values for my test instances.

I have run the model on the windows server with 32 processors and 128 GB memory. However, I could not identify a significant difference in terms of solution quality and run-time, when I compared the server against my laptop (i-5 single cpu and 8 GB). I guess I am missing something. Only parameter I am aware of is # of Workers (cp.param.Workers = 32 on server). Is there any parameter I need to configure to utilize the luxury of multi-cpu and large-memory on the server?

Regards,
Andy

Hi,

just to check, have you tried to see the impact of random seed ?

https://www.ibm.com/developerworks/community/forums/html/topic?id=04737d2f-6a82-4707-84f0-85807130ca98&ps=25

regards

Originally posted by: AndyHam

Wow. This is awesome ^^
I plugged my model in your code.
It shows a very high variation in obj values.
I think it will be perfect if I can assign each random seed to each of 32 CPUs.
Then, report the best from the 32 samples.
Can you please teach me how to accomplish this goal?

Hi,

we have a specific docplexcloud forum at https://developer.ibm.com/answers/smart-spaces/94/decision-optimization.html

What you could do is within the IDE have 32 run configurations with nbWorkers=1 and different random seed and then run them in // from the IDE

Or rely on https://developer.ibm.com/answers/questions/371449/how-to-import-docplexcloud-solution-into-a-local-o/?smartspace=decision-optimization

and call 32 jobs in // from Python

Or you stay serial and run them one after the other one as you did previously.

regards