A Simple Adaptive Cache Algorithm Outperforms LRU

Discarding the least recently used (LRU) item is the policy of choice in cache management. Until recently, attempts to outperform LRU in practice had not succeeded because of overhead issues and the need to pre-tune parameters. This article presents a new self-tuning, low overhead algorithm, namely, Adaptive Replacement Cache (ARC). It responds online to changing access patterns, continually balances between the recency and frequency features of the workload, and demonstrates that adaptation eliminates the need for workload specific pre-tuning. Like LRU, ARC can be easily implemented and its running time per request is essentially independent of the cache size. Also, unlike LRU, ARC is "scan-Resistant" in that it allows one-time sequential requests to pass through without polluting the cache. ARC leads to substantial performance gains over LRU for a wide range of cache sizes. For example, on a workstation disk drive workload, at 16MB cache, LRU delivers a hit ratio of 4.24% while ARC achieves a hit ratio of 23.82%, and, for a SPC1 benchmark, at 4GB cache, LRU delivers a hit ratio of 9.19% while ARC achieves a hit ratio of 20%.