Deep Reinforcement Learning Based Edge Computing Offloading Mechanism

Abstract

With the development of the end users, the cloud computing service architecture is no longer suitable for low latency. At present, a large amount of electric and computing power is wasted. In this paper, we propose a two-stage task offloading strategy to ensure the quality of service while minimizing energy consumption. We use Google cluster data[l] and consider the dependencies between different tasks and parallel processing to build user workload model. The mobile edge computing(MEC) environment includes cloud environment server and base station processing unit. A two-level processor based on deep Q-learning is used to automatically generate the best long-term decision. In order to evaluate the effectiveness of the model, we use the sequential allocation method as the baseline, and compare the running time, with energy consumption and failure rate as the measurement indicators. Based on the test, the average energy consumption of our system is increased and the failure rate is reduced about 40% compared with the baseline.