Moving Server: Follow-up Computation Offloading Paradigm for Vehicular Users

With the emergence of ever-growing compute-intensive applications, geo-distributed cloud resources are deployed at the mobile edge and alleviate the resulting surging demand in communication and computation. For a high-velocity vehicular user, however, frequent handovers among wireless base stations and corresponding task migrations among edge servers can lead to throughput degradation, longer latency, and increased energy consumption. To tackle this issue, we design a novel follow-up computation offloading paradigm, where moving servers can provide extra computing resources and the deep Q-learning based computation offloading strategy will improve the quality of service. Our study provides insights into the following questions: Who is capable of being a moving server? How to offload to a moving server?, and Which server is the best choice? Furthermore, to resolve the conflict between the long convergence time of learning algorithm and high mobility of the vehicular user, we enhance the server selection scheme by incorporating the prior probability of the availability of each moving server. The simulation results demonstrate the efficiency and performance superiority of our proposed paradigm over the existing schemes.