Comparing Methods

With the availability of parallel computers the question how to compare sequential and parallel algorithms is an interesting question.

Some algorithms parallelizes more nicely than others, i.e., the overhead from communication between the threads differs and they experience different rates of computation that cannot be parallelized. At best there is no overhead and then using k processors in parallel is the same as running the same computations k times in sequence on a single processor computer.

In such a case it seems not fair to say that the parallel algorithm was much superior to the sequential one, but rather that their preformance was equal, because k times more resources were used by the parallel algorithm.

Parallel algorithms should therefore be compared on their own.

Most comparisons of methods in publications on global optimization do not honor fair comparison in the sense presented above and therefore the conclusions reached can be questioned.