Q-learning based collaborative cache allocation in mobile edge computing

Abstract The rapid development of  Augmented Reality (AR), Virtual Reality(VR), Internet of Things (IoT), and high-definition video has caught the attention of low latency and high bandwidth network requirements. For huge data transmission, cache technology has been regarded as an effective solution for reducing transmission time from users to remote clouds. However, the increasing variety of cached data has caused the challenge of cache performance. Based on high flexibility, scalability, and deployability of the characteristics of Software-Defined Networking (SDN), this study proposes a collaborative cache mechanism in multiple Remote Radio Heads (RRHs) to multiple Baseband Units (BBUs). In addition, the traditional rule-based and metaheuristics methods are difficult to consider all environmental factors. To reduce the traffic load of backhaul and transmission latency from the remote cloud, we use Q-learning to design the cache mechanism and propose an action selection strategy for the cache problem. Through reinforcement learning to find the appropriate cache state. The simulation results show that the proposed method can effectively improve the cache performance.