Re: [Starpu-devel] Equivalence vs substitutability

[Samuel Thibault <samuel.thibault@ens-lyon.org>]

Hello,

Chris Hennick, le Tue 09 Jul 2013 10:06:58 +0200, a écrit :
> the two implementations would probably be subject to separate
> diminishing returns, so each would have an upward trend in running time, and
> extrapolating that trend would give a better estimate of running time than
> using a long-term average.
> 
> Does StarPU have any way to estimate running times in this situation?

Not with the current performance models.  An additional model could
however be written that merely remembers the last e.g. 10 measures, and
make an average over these.

> I don't
> think I've seen any research on task scheduling where it was acknowledged 
> that
> processes could be substitutable without being black-box equivalent;

That's a tricky thing :) Turning the tasks into something comparable,
like you suggested, makes it indeed way easier to take into account for
the scheduler.

Samuel