next up previous
Next: Cache Size Up: Simulation Previous: Maximum Order of

Improvements Over LRU

With a firm understanding for the appropriate parameters of our model, we compared our predictive cache to an LRU cache. For this comparison, both caches simulated four megabytes of cache memory. Our predictive cache extended to the second order and prefetched at a conservative threshold of 0.1. Table 1 shows the results of our simulations. Our predictive cache clearly offered significant improvements to the performance of LRU on all eight traces, averaging 15% more cache hits than LRU and in the case of trace E as much as 22% more.



Tom M. Kroeger Tue Nov 14 00:27:35 PST 1995