Analysis of Similarity Caching on General Cache Networks

Abstract

Nowadays, content caching is essential to serve contents to users quickly and efficiently. In recent years, a technique to improve content caching, similarity caching, which focuses on the similarity among contents, has emerged. The feature of similarity caching is that a cache provides any content similar to a user’s request even though the cache does not have the request itself, i.e., cache miss. In the literature, the effectiveness of similarity caching is investigated mainly through system-level metrics such as the cache hit ratio. However, its effectiveness from the point of view of users remains hardly understood despite its importance. Therefore, we try to quantify the user-level performance of similarity caching while focusing on a general cache network comprised of multiple cache nodes. To this aim, we introduce a user-oriented metric called similar-content delivery delay and analytically derive that; we also analyze the relation between the delivery delay observed by a user and the utility of the content similarity through several numerical evaluation. Our key findings are summarized as follows; (i) the effectiveness of similarity caching is dominated by the similarity which is acceptable to users; (ii) the effectiveness is also dependent on the location of users, i.e., distance to a contents server.

Keywords

• Content distribution networks
• information-centric networking
• mathematical analysis
• quality of service