Pmap structure master process computes too

Specific Domains Julia at Scale
1

What does happen if the master process handles the part part after remotecall_fetch but computes itself as well? Does everything work or are there some catches?

I have two pmap like functions nested and the first calls the processors 2,4,6,… and each of them calls two procs well. What I currently do in the inner one is that processor 2 calls processor 2 and 3 in the @async part so it has to handle the calls and computes itself. (The handling is a quite small part) My question is: Should this be avoided?
I’m not sure about just create new procs for it as the user specifies how many processors are used and it feels a little bit like cheating if I introduce new procs especially for benchmarking.

2

One problem I see is that when I run remotecall_fetch and use the current proc then it can just call that function and when it’s done continue → the second proc is not really used?
Which should be resolved when using all other processors beforehand and the master is the last that gets called?