Dispatching and Scheduling Dependent Tasks Based on Multi-agent Deep Reinforcement Learning

With the development of edge computing, a large number of tasks can be offloaded to the edge server for computing, among which the dispatching and scheduling of dependent tasks has attracted extensive attention. The offloading of dependent tasks mainly has the following problems: how to select an appropriate edge server for dispatching, how to arrange the scheduling order of edge servers to better schedule tasks, and how to solve the task dependency problem. In this paper, we proposal a dispatching and scheduling method DAMD, based on reinforcement learning and multi-agent reinforcement learning, to solve the above three problems. Specifically, as the first step of DAMD, a reinforcement learning approach is designed to estimate the network load and dynamically dispatch tasks to the appropriate edge servers. Each edge server is regarded as an agent by a multi-agent reinforcement learning method, the second step of DAMD, which comprehensively considers the dependency relationship between tasks and the scheduling relationship between servers to achieve the efficiency and fairness of task scheduling. Finally, the results show that our method can better complete the task within the deadline and greatly reduce the average response time according to the time sensitivity requirement.