I think the speedup should be defined as runtime of fastest sequential program / runtime of parallel program on multiple processors
Nesuna
To elaborate on the above point, the idea is that if you're pouring a lot of effort into making this parallel program run well, it only makes sense to compare it to an equally "effortful" attempt at the sequential algorithm since running things in parallel consumes more resources that, if the sequential algorithm runs fine, then you would want to conclude that you don't have that much speedup using the parallel algorithm.
flxpox
Isn't that in most cases running the parallel version of the program on one processor equivalent to the best sequential program? Just because the parallel version of the program would parallelize as much as possible and left the necessary sequential part as it is.
Where Last week is.
I think the speedup should be defined as runtime of fastest sequential program / runtime of parallel program on multiple processors
To elaborate on the above point, the idea is that if you're pouring a lot of effort into making this parallel program run well, it only makes sense to compare it to an equally "effortful" attempt at the sequential algorithm since running things in parallel consumes more resources that, if the sequential algorithm runs fine, then you would want to conclude that you don't have that much speedup using the parallel algorithm.
Isn't that in most cases running the parallel version of the program on one processor equivalent to the best sequential program? Just because the parallel version of the program would parallelize as much as possible and left the necessary sequential part as it is.