Enhanced edge offloading using Reinforcement learning

Internet of Things (IoT) based solutions requiring real time results from intensive computation tasks or having large scale data analysis have traditionally been designed with offloading of the work to cloud infrastructure. This has been found to be not an ideal solution due to several issues related to network uncertainties, cost of cloud usage etc. This is especially true for systems with both hard time constraints and large amount of data. Edge computing, with its hierarchical configuration has been proposed to solve these issues. This has led to researchers proposing several algorithms to optimise offloading of computation to the layers of this hierarchy. In this work we propose the use of an actor-critic based reinforcement learning mechanism to solve the offloading planning for a general hierarchical system with multiple end nodes and multiple edge servers. Our simulation based results shows that the proposed method improves the performance of the system as compared to the existing benchmark offloading policies.