Scope and method of study. The purpose of this study was to examine the efficacy of modifying a hardware cache prefetcher to track and predict context switches and prefetch for incoming processes before they are switched in. The study was composed of three major components--quantifying the amount of context switches that can be correctly predicted, examining out of context prefetching on CPU-bound processes and I/O-bound processes, and examining the effects of varying how early before a context switch to begin prefetching.; Findings and conclusions. Data suggests that highly accurate context switch prediction is viable, with our own simple prediction unit correctly predicting over 75% of context switches. The study shows that out of context prefetching may not work well with CPU-bound processes, as the positive effects are masked by the relatively long timeslice lengths. Finally, performing out of context prefetching on I/O-bound processes exhibits significant performance improvements over traditional cache prefetching.
展开▼
