A Study of the Reuse-Distance Behavior of Instructions

Abstract: Cache replacement policies play a critical role in optimizing the performance of cache memory in computer systems. While policies like Least Recently Used (LRU) are popular, they have proven to be inefficient under certain circumstances. This master's thesis investigates the concept of reuse distance as a promising approach to predict the future usage of cachelines for optimal cache replacement. Reuse distance refers to the number of unique memory accesses between two successive accesses to the same memory location. Previous work introduced a technique for predicting reuse distances based on load/store instruction sequences, but it has not been applied to the reuse distances of instructions in the instruction cache. This research fills this gap by conducting a comprehensive examination of the reuse distances of instructions. In doing so, it aims to explore the extent of any disparities between the reuse distances of instructions and data, and the implications this may have on cache replacement strategies. The ultimate objective is to contribute to the development of more efficient cache replacement algorithms by incorporating insights on instruction reuse distances.