Effect of speculative prefetching on network load in distributed systems

Abstract

Previous studies in speculative prefetching focus on building and evaluating access models for the purpose of access prediction. This paper, on the other hand, investigates the performance of speculative prefetching. When prefetching is performed speculatively, there is bound to be an increase in the network load. Furthermore, the prefetched items must compete for space with existing cache occupants. These two factors-increased load and eviction of potentially useful cache entries-are considered in the analysis. We obtain the following conclusion: to maximise the improvement in access time, prefetch exclusively all items with access probabilities exceeding a certain threshold.