무선 엣지 네트워크에서 DQN을 활용한 캐시 콘텐츠 교체

Abstract

According to the recent report, global mobile data traffic is expected to be increased by 11 times over the next five years. Moreover, the growth in mobile data traffic is expected to be driven mainly by mobile video traffic. The video traffic is expected to account for about 70 percent of the total of mobile data traffic. To solve these enormous mobile data traffic's problem, we need to understand video traffic's characteristic. Recently, the repetitive requests of some popular content such as popular Youtube video clips cause a enormous network traffic overheads. If we constitute a wireless edge network with the nodes capable of content caching based on the content popularity, we can reduce the network overheads by using the cached content instead of receiving it from a base station for every request. In wireless edge networks, caching is attracting attention as a way to mitigate backhaul link congestion and efficiently handle the massive growth of mobile video traffic. Content placement / replacement is an important issue to maximize caching gain with limited storage space, but it is difficult to get effective answers with existing optimization techniques. In this paper, we present a caching strategy using reinforcement learning to solve these problems, and discuss the performance gain in cache placement / replacement problems. We introduce the problems of caching using q-learning, propose a caching strategy using Deep Q-Network to overcome them, and look at the improved benefits.