FRD: A Filtering based Buffer Cache Algorithm that Considers both Frequency and Reuse Distance

Buffer cache algorithms play a major role in filling the large performance gap between main memory and I/O devices in a mass storage system. Many buffer cache algorithms have been developed such as the low inter-reference recency set (LIRS) and adaptive replacement cache (ARC). Careful analysis of real-world workloads leads us to observe that approximately 50 to 90% blocks are accessed three or fewer times during the execution of various workloads. These infrequently accessed blocks are likely to cause high cache pollution by evicting better blocks from the cache. We also observe that these infrequently accessed blocks have certain access characteristics regarding reuse distance: either extremely long or short. Based on our observations, we propose an algorithm named frequency and reuse distance (FRD). The proposed algorithm concentrates on the manner in which to utilize both the access frequency and reuse distance of a block to determine the entries that must be stored in a buffer cache. We implemented the FRD algorithm using two LRU stacks. Experimental results show that the proposed algorithm outperforms state-of-the-art cache algorithms such as LIRS and ARC in most cases. We also show that FRD’s hit ratio is stable under various cache sizes. Keywords—buffer cache; reuse distance; frequency