UC Santa Cruz

Prefetching multiple files per prediction can improve the predictive accuracy. However, it comes with the cost of using extra cache space and disk bandwidth. This pa- per discusses the most Recent distinct 􏰋 Successor (RnS) model and uses it to demonstrate the effectiveness of our earlier work, Program-based Last 􏰋 Successor (PLnS) model, a program-based prediction algorithm [21]. We analyze the simulation results from different trace data and show that PLnS can perform better than RnS while it only predicts at most 59% of the number of files pre- dicted by RnS when the 􏰋 in PLnS equals to two. PLnS is a good candidate when considering prefetching multiple files per prediction to improve predictive accuracy.