[mpich-discuss] MPI_Probe

[Nicolas Rosner]

> If you really care, though, you could always do the polling yourself
> (using Iprobe instead of Probe), couldn't you?
> But it would surprise me if such a decision paid off merely
> due to being able to fine-tune the polling freq.

I meant that in a context like the one you describe, where reaction
time still does matter. Achieving some decent compromise between that
and wasted resources is what implementations try to do, I'm guessing.

Of course, if instead one knew e.g. that for some app any desirable
freq needs to be hundreds of times lower than a reasonable
general-purpose compromise, the above could make a lot of sense. But
is that the case here? You seem to prefer a balanced choice as well.

Cheers, N.