A multi-edge jointly offloading method considering group cooperation topology features in edge computing networks

Abstract

Mobile Edge Computing (MEC) is a new computing paradigm that has shown great potential. How to extract the cooperative topological relationship between MEC servers to realize jointly computing is the key problem to solve the bottleneck of MEC computational capability. In previous studies, multi-MEC servers are regarded as unit computing nodes with the same cooperation relationship to jointly schedule offloading tasks, without considering the hierarchical and clustered topology of the server collaborative work. As a result, in the scenario of unbalanced distribution of computing resources, it is difficult to obtain the optimal joint scheduling strategy for offloading tasks according to the cooperation relationship and resource differences among MEC servers. Therefore, this paper considers introducing the topological relationship of group cooperation among multi-MEC servers to optimize the joint scheduling strategy, and proposes a Multi-Agent Hierarchical Graph Attention Soft Actor-Critic algorithm (MHSAC). Firstly, based on the differences in their own resources and the demands of the tasks they undertake, MEC servers are divided into series clusters. Then, a Hierarchical Graph Attention Network (HGAT) is used to model each agent to extract the physical communication topology information of the MEC server and the group topology information of multi-edge cooperation. The multi-agent soft Actor-Critic algorithm is used to obtain the offloading scheduling decision of multi-edge cooperation. Experiments show that the MHSAC algorithm that considering the topological relationship of multi-edge group cooperation can optimize load distribution under low latency and resource-limited requirements, achieving optimal load balancing values and task drop rates.