Number of tasks that can run on a single node

Ray Core
1

I want to understand better how many tasks Ray can schedule at a time on a single node.

Currently I am declaring an actor like this:
Actor.options(num_cpus=0.01, resources={“node:” + ip : 0.01}

Then I proceed to invoke X invocations of a remote method of Actor, on a machine with N CPUs. Will they be scheduled to run all at the same time? (I want to overload the CPUs of my machine, and launch way more tasks than there are cores)

2

Will they be scheduled to run all at the same time?

Actor can process 1 task at a time by default. You can use multi processing here, but Python has GIL, so you are essentially invoking 1 task at a time if you send tasks to the actor using actor methods.

If you’d like to utilize multiple CPUs, you should start many actors and send tasks to all over them (you can also use Ray’s ActorPool), or you can use Ray tasks instead.