Knowledge-centric proactive edge caching over mobile content distribution network

Abstract

By enabling caching content at base stations, accessing points or other mobile nodes which are close to users, wireless edge caching is able to shorten the content delivery latency and offloading data traffic from backhaul to local, which has become a promising solution video service in future mobile content distribution network. How to maximize the caching utilization while averagely reduce the overall video delivery latency is a critical issue in wireless edge caching. However, the high dynamic and complex of user playback behavior of mobile video streaming makes it difficult to decide which content should be cached by popularity investigation or conditional probability-based requests prediction. In this paper, we propose a knowledge-centric edge caching optimization for multimedia service in 5G that online optimizes the caching configuration based on semantic information of user playback behavior. We first capture the semantic information of user by employing a deep belief networks. Then, by determining the future requested video based on playback pattern, we mathematically formulate the caching optimization problem which is NP-complete but submodular and monotone, namely a near optimal solution can be found in polynomial time. To solve this optimization problem, we propose a greed-based online caching configuration algorithm in order to minimize the overall delivery cost for video streaming in real-time. Simulation results show how our proposed method outperform the state-of-the-art solutions.