Amdahl’s Law is often used in parallel computing to predict the theoretical maximum speedup using multiple processors. Now, Future Chips looks at cases where Amdahl’s Law doesn’t apply:
“As with any analytic model, Amdahl’s law should only be used when a workload fits the programming model assumed by the model. There are many programs that do not fit this model. Fundamentally, Amdahl’s law assumes that any code which cannot be parallelized is always on the critical program path. This is not the case in many modern programming paradigms as some non-parallel code sections can run in parallel with other independent code.
Read the Full Story or check out a recent interview with Gene Amdahl over at the Massively Parallel site.
I hope people don’t read the article and think that it’s a real criticism of Amdahl. the example it gives (with multiple phases and varying levels of parallelism) is really a reiteration of Amdahl’s asymptotic analysis. it really is all about the serial fraction, since you can only reduce the parallel sections to zero cost…