Hello,
I'm working on distributed reinforcement learning project. I've observed that simulations are slowing down with new instance spawned. For examples, for one instance sim step is 60ms, for two instances sim step in both is 80ms, etc. The computer on which I'm running experiments has Ryzen 9 3900X 12c/24t. I'm using PyRep lib, but same effect was observed with custom Remote API lib.
Could you give me any advice how to eliminate this effect?
Greg.
Slow down with multiple instances
Re: Slow down with multiple instances
Hello Greg,
it would be an ideal world if each additionally spawned instance wouldn't have any effect on the rest of the system. Unfortunately this is not the case. The OS is free to allocate core threads and other resources (GPU) as it wants. So it could be that the second instance is sharing the same thread(s) as the first instance.
There is unfortunately not much that you can do, except maybe give some preference hints to the OS, whether this is feasible or not is another question.
Cheers
it would be an ideal world if each additionally spawned instance wouldn't have any effect on the rest of the system. Unfortunately this is not the case. The OS is free to allocate core threads and other resources (GPU) as it wants. So it could be that the second instance is sharing the same thread(s) as the first instance.
There is unfortunately not much that you can do, except maybe give some preference hints to the OS, whether this is feasible or not is another question.
Cheers