Issue No. 04 - April (2004 vol. 37)
Dharmendra S. Modha , IBM Almaden Research Center
Nimrod Megiddo , IBM Almaden Research Center
The policy of choice for managing cache memory has been to discard the least-recently-used item. LRU has consistently outperformed other policies by virtue of its simplicity and low overhead. However, an adaptive replacement cache algorithm offers a low-overhead alternative that shows substantial performance gains over LRU. ARC's online adaptation continually balances between a workload?s recency and frequency features, eliminating the need for the workload-specific pretuning that plagued many previous proposals to improve LRU. The algorithm has the additional advantage of being scan-resistant so that one-time sequential requests can pass through without polluting the cache.
Dharmendra S. Modha, Nimrod Megiddo, "Outperforming LRU with an Adaptive Replacement Cache Algorithm", Computer, vol. 37, no. , pp. 58-65, April 2004, doi:10.1109/MC.2004.1297303