Experimental analysis of caching efficiency for YouTube traffic in an ISP network

Abstract

In this paper, we report YouTube traffic measurements from Orange IP backbone network connecting residential customers. We exhibit its salient features in relation to the performance of caching. By examining the file popularity distribution, we show that video requests are highly volatile in that a huge number of files are viewed only a few times; these files are therefore not relevant for caching. Nevertheless, there is a subset of files which are massively viewed by end users and are worth caching. On the basis of this experimental observation, we develop a mathematical model for estimating the efficiency of file caching in the presence of noise traffic composed of those files which are rarely requested and thus “pollute” the cache. We then proceed to trace-driven simulations in order to check the qualitative conclusions derived from the theoretical model.