A case study of larc admission control on web caching.

Abstract

Caching is crucial for improving the performance of Web services. The key factor that determines the quality of caching is its capacity. If the cache capacity is too small, caching performance (e.g. hit ratio) is impacted. However, one cannot simply add more capacity to improve performance. If cache capacity grows beyond a certain point, caching performance is not improved, and the added capacity is wasted. For a deined capacity, caching systems apply a replacement policy that decides which items must be removed when the cache is full. The literature has a lot of cache replacement policies. In this work, we take another direction. We analyze how to be thrifty using cache capacity based on admission control policies. Instead of deciding which items should be removed from the cache, the admission control decides whether an item should enter the cache. We considered LARC, a well-known admission policy that allows items to be cached only in their second access. We evaluated this cache policy by simulating a request trace of a large e-commerce web service. The results indicate that when the cache is oversized, e.g. its capacity is larger than the optimal, the admission control policy reduces the cache usage up to 50%, with a small 5% performance penalty. On the other hand, when the cache is undersized, the admission control policy increases the cache performance up to 22% with no cost on cache usage increase. We believe these results indicate that there is potential for integrating admission control into cache management.