Learning-based Content Caching in Collaborative Edge Networks

Wireless caching techniques are widely adopted in edge computing systems to improve the quality of services at the edge of the network. This paper studies the problem of online content caching and elimination without the in-advance knowledge of content popularity in a collaborative edge caching network, which consists of a set of edge nodes equipped with base stations and storage units. We aim to minimize the content providers’ cost while ensuring the quality of service (QoS) in the long run under the uncertainties of requested content. To this end, we formulate the problem of minimizing the system cost consisting of cache cost, migration cost and QoS penalty cost. In order to address this problem, we propose an algorithm for Collaborative Edge Caching based on the Thompson Sampling (CEC-TS) technique to update the caching decisions via adaptive learning of the popularity of requested contents. Simulation results demonstrate that the proposed CEC-TS algorithm achieves lower content providers’ and QoS penalty cost against baseline schemes, i.e., the cloud-only and the first-in first-out (FIFO), in the long run.