Resource-Aware Microservice Deployment in Cyber–Physical–Social Systems via Distributed Intelligence

Microservice-based architecture has become prevalent in the development of diverse types of applications and systems within cyber–physical–social systems (CPSSs), leading to the development and deployment of microservices that provide a range of functionalities in CPSSs. An effective deployment solution for microservices holds promise for enhancing response times in CPSSs. However, prior research typically overlooks several important elements, for example, the influence of call relationships among microservices, and the various challenges introduced by network environments in edge networks. In this article, to address those issues, we propose to represent the call relationships between microservices using an undirected graph, and our motivation is to effectively obtain the communication overhead resulting from the physical separation of servers in edge environments. For edge networks, we fully considers the typical instability factors such as packet loss, delay, and bandwidth. We develop a multiobjective optimization task for microservice deployment in edge networks, where the object is minimizing communication overhead while accommodating unstable network conditions. To better generate the final deployment strategy, we develop an innovative reinforcement learning (RL) algorithm that is capable of generating the Pareto front set for multiobjective optimization and selecting the optimal solution from this set. Extensive experiments were performed on a real-world distributed computing platform, evaluating a set of objects (e.g., response time, overhead, loss, and delay) in many request volume cases. The results demonstrate that our algorithm achieves the best performance in all test cases compared to alternative methods and the default strategies of Kubernetes.