Content-centric Caching Using Deep Reinforcement Learning in Mobile Computing

In era of Internet, the amount of the connected devices has been remarkably increasing along with the increment of the network-based service. Both service quality and user's experience are facing great impact from latency issue while a large volume of concurrent user requests are made in the context of mobile computing. Deploying caching techniques at base stations or edge nodes is an alternative for dealing with the latency time issue. However, traditional caching techniques, e.g. Least Recently Used (LRU) or Least Frequently Used (LFU), cannot efficiently resolve latency caused by the complex content-oriented popularity distribution. In this paper, we propose a Deep Reinforcement Learning (DPL)-based approach to make the caching storage adaptable for dynamic and complicated mobile networking environment. The proposed mechanism does not need priori knowledge of the popularity distribution, so that it has a higher-level adoptability and flexibility in practice, compared with LRU and LFU. Our evaluation also compares the proposed approach with other deep learning methods and the results have suggested that our approach has a higher accuracy.