The least semantically related cache replacement algorithm

Abstract

This article presents the Web cache replacement algorithm named Least Semantically Related (LSR), as an alternative to well-known and widely employed replacement policies, such as SIZE, LFU and LRU, which are based on physical properties of objects (documents). LSR is based on the semantics of the information contained within objects: LSR tends to evict objects which are less related to the last object to enter the cache with respect to their semantics, therefore preserving objects of more interest to clients. A detailed algorithm and a data structure for LSR where objects are classified according to a previously established taxonomy were designed and implemented for the purposes of validation and comparison with other replacement policies. Besides, a framework for simulation was designed and verified, including the data preparation process. The initial experimental results show that LSR out-performs traditional policies in terms of hit rate, in most cases.