Multi-Armed Bandit-based Routing Method for In-network Caching

This paper proposes a routing method based on the multi-armed bandit algorithm for in-network caching. In-network caching is a technology that enables intermediate routers to store contents in their cache. Thus, users can download the contents from the routers in addition to original content servers. This feature can save network resources and reduce response time. The proposed method applies multi-armed bandit algorithm, which is one of reinforcement learning, to route selection of the routers, where we consider the routers as players in a multi-armed bandit problem. In the proposed method, each router forwards content requests to appropriate output ports based on rewards calculated by the multi-armed bandit algorithm. Through simulation experiments, we show that the cache hit rate is improved by about 10 % while suppressing the increase in the average hop counts, compared with a shortest path algorithm.