Comparison of multi-armed bandit algorithms for content request routing in cache-enabled networks

In cache-enabled networks, contents are cached in intermediate routers to be downloaded by clients. In the past, a multi-armed bandit (MAB) problem-based routing (named MAB-routing) has been proposed in order to efficiently utilize the cached contents. In the MAB-routing, each intermediate router forwards an arrival content request to an appropriate output port based on an MAB algorithm, regarding output ports as arms. By doing so, the MAB-routing becomes increasingly likely to find the corresponding content cached in a router. In this paper, we examine how MAB algorithms affect the performance of the MAB-routing. There exist several MAB algorithms such as epsilon-greedy, Upper Confidence Bound, and Thompson Sampling. Through simulation experiments, we show the performance of the MAB-routing with each algorithm.