Reinforcement Learning-Based Routing Method Utilizing Hierarchical Content Names for In-Network Caching
Ippei Tayuki, Kouji Hirata · IEEJ Transactions on Electronics Information and Systems · 2025
In-network caching enables us to efficiently retrieve contents by storing the contents in the cache storage of routers. In order to utilize in-network caching environments, an appropriate routing method for content requests is required. This paper proposes a reinforcement learning-based routing method using hierarchical content names for in-network caching. The proposed method applies the multi-armed bandit (MAB) algorithm, which is one of reinforcement learning, to routing decision at routers. The routers have tree-shaped routing tables managing hierarchical names of contents. Content requests arriving at the routers are forwarded to directions determined by the MAB algorithm referring to the hierarchical content names. The proposed method dynamically changes the tree-shaped routing tables according to arrival content requests, so that it can efficiently reduce the size of routing tables. Through simulation experiments, we show that the proposed method keeps high cache hit ratio and low hop counts to retrieve contents, while reducing the size of routing tables.