Role of aging, frequency, and size in web cache replacement policies

Abstract. Document caching on is used to improve Web performance. An eÆcient caching policy keeps popular documents in the cache and replaces rarely used ones. The latest web cache replacement policies in-corporate the document size, frequency, and age in the decision process. The recently-proposed and very popular Greedy-Dual-Size (GDS) policy is based on document size and has an elegant aging mechanism. Simi-larly, the Greedy-Dual-Frequency (GDF) policy takes into account le frequency and exploits the aging mechanism to deal with cache pollution. The eÆciency of a cache replacement policy can be evaluated along two popular metrics: le hit ratio and byte hit ratio. Using four dierent web server logs, we show that GDS-like replacement policies emphasizing size yield the best le hit ratio but typically show poor byte hit ratio, while GDF-like replacement policies emphasizing frequency have better byte hit ratio but result in worse le hit ratio. In this paper, we propose a generalization of Greedy-Dual-Frequency-Size policy which allows to bal-ance the emphasis on size vs. frequency. We perform a sensitivity study to derive the impact of size and frequency on le and byte hit ratio, identifying parameters that aim at optimizing both metrics.