Deep Reinforcement Learning Based Container Cluster Placement Strategy in Edge Computing Environment
Zhuo Chen, Bowen Zhu · GLOBECOM 2022 - 2022 IEEE Global Communications Conference · 2022
Container-based virtualization technology has been increasingly used in edge clouds recently due to its advantages of lighter resource occupation, faster startup capability, and better resource utilization efficiency. For a task request, it is usually necessary to interconnect multiple containers to build a Container Cluster (CC), and then deploy it to edge service nodes with relatively limited resources. However, the increasingly complex and time-varying nature of tasks brings great challenges to optimally deploying CC on edge nodes. From the perspective of edge service provider, this paper regards the charges for various resources occupied by providing services as revenue, and regards the service efficiency and energy consumption generated by providing services as cost, then formulate an integer optimization model to describe the optimal deployment of CC on distributed edge nodes. Different from the heuristic-based methods adopted in existing works, we introduce Graph Convolutional Network (GCN) to extract features from the logical topological link relationships among containers in CCs and use them as the input of the solution framework to improve the solution quality. Compared with other related algorithms, the results show that the method proposed in this paper can obtain better solution quality and maintain acceptable solution efficiency under the same edge cloud environment. The improved system revenue for edge service provider can be further obtained.