Limits on the performance benefits of multithreading and prefetching

This paper presents new analytical models of the performance be-nefits of multithreading and prefetching, and experimental nleasure-ments of parallel applications on the MIT Alewife multiprocessor. For the first time, both techniques are evaluated on a real machine asopposedto simulations. Themodels determine theregion in the parameter space where the techniques are most effective, while the measwements detemine theregion where the applications lie. We find that these regions do not always overlap significantly. The multithreading model shows that only 2A contexts are ne-cessary to maximize this technique’s potential benefit in current multiprocessors. Multithreading improves execution time by less than 10 % for most of the applications that we examined. The model also shows that multithreading can significantly improve the per-formance of the same applications in multiprocessors with longer latencies. Reducing context-switch overhead is not crucial, The software prefetchingmodel shows that allowing 4 outstand-ing prefetches is sufficient to achieve most of this technique’s poten-tial benefit on current multiprocessors. Prefetehing improves per-formance over a wide range of parameters, and improves execution time by as much as 20-50 % even on current multiprocessors. The two models show that prefetching has a significant advantage over multithreading for machines with low memory latencies and/or ap-plications with high cache miss rates because a prefetch instruction consumes less time than a context-switch.