[R-sig-hpc] Snow, parApply computational times

Elena Grassi grassi.e at gmail.com
Fri Sep 28 16:28:21 CEST 2012


On Fri, Sep 28, 2012 at 3:22 PM, Martin Morgan <mtmorgan at fhcrc.org> wrote:

> the problem is that the amount of 'work' for each calculation is small, so
> that any gain from parallel calculation is offset by the cost of sending
> data back and forth, etc.

Yup, that's a sensible explanation, even if having only a single
process alive for the majority
of the computational time was an hint of something slightly different
from my point of view, maybe wrong :)

Thank you very much for your suggestion, I will give it a try right
away (as soon as our computational resources are free, at least :) ).

E.



More information about the R-sig-hpc mailing list