Graph Convolutional Network Augmented Deep Reinforcement Learning for Dependent Task Offloading in Mobile Edge Computing

In this paper, we study the problem of dependent task offloading in mobile edge computing. Applications running on mobile devices require computing, but the computing resources of the mobile device are too few to meet the demand. To solve this problem, mobile devices can offload applications to nearby edge nodes, which are devices equipped with computing resources, for execution. Typically, each application can be divided into a set of dependent tasks whose execution dependencies form a directed acyclic graph (DAG). How to optimally offload these dependent tasks to resource-constrained edge nodes with the least application completion time (defined as the makespan) is a challenge. In this paper, we address this problem and propose a novel solution called GRL, which is a Deep Reinforcement Learning (DRL) model augmented by Graph Convolutional Networks (GCN). The rationale for our design is to utilize GCN to describe task dependencies and transform the problem of dependent task offloading into a node classification problem, thus perfectly handling non-Euclidean data. Simulation results show that the GRL model outperforms other algorithms in terms of the makespan of the application. It performs very well and converges quickly on various DAG structures of applications, even on real-world datasets.