부산대학교 도서관

Edge computing is the complement of cloud computing which provides low-latency and real-time responsive cloud-like services at the edge of the network. Compared with the abundant resources of cloud computing, edge servers have limited resources and heterogeneous, which brings challenges to efficient service placement for multiple mobile users in edge computing. Moreover, the diversity of end devices and the mobility of users also make the placement problem complicate. This paper studies the service placement problem to joint optimize the utilization of edge servers and reduce users' latency. We first introduce a novel multi-level master-slave architecture based on the cloud-edge-device scenario. Based on that, we propose a framework to solve the service placement problem for multi-mobile users. In this framework, we mitigate edge system resource differences and multi-user trajectory unpredictability problems through service feature analysis and user trajectory prediction, respectively. Combining the results of the feature analysis and trajectory prediction, we propose an active service placement method based on deep reinforcement learning. This paper conducts experiments in a simulation environment based on real data sets, and the results show that the algorithm can effectively reduce users' delay under the condition of optimizing edge server utilization.