Computation Offloading in Edge Computing Based on Deep Reinforcement Learning

The development of 5G technology provides great convenience for edge computing. But it also means that the computing tasks generated by user equipment are more and more complex. These tasks are no longer simply to complete one target, but are often composed of multiple subtasks. In order to solve the edge computing problem of multiple subtasks, we propose a Task Mapping Algorithm based on Deep Reinforcement Learning (DRL-TMA). Firstly, we abstract the computation intensive task as a directed acyclic graph, and then propose a Graph Sequence Algorithm (GSA) to transform the DAG task into a specific topological sequence, and then determine the optimal offloading decision of all subtasks according to the sequence order. We model the offloading problem as a Markov Decision Process (MDP) to maximize the comprehensive profit. The experimental results show that the Task Mapping Algorithm based on Deep Reinforcement Learning has stronger decision-making ability and can obtain the approximate optimal comprehensive profit which proves the effectiveness of the algorithm.