MAPPO-based Predictive Auto-Scaling Algorithm in Edge Cloud Clusters

In edge cloud clusters, application instances are often deployed on edge nodes close to users, thereby reducing response time. However, the computing capacity of edge nodes is limited, necessitating the use of auto-scaling technology to dynamically adjust the number of application instances on the nodes based on the load. In this paper, we propose a predictive auto-scaling algorithm to improve the cluster resource utilization efficiency. Specifically, the service workload of each edge node is predicted by a time series parallel prediction model. Then, based on these prediction results of service workload, the autoscaling decisions of edge nodes for resource adjustment in each cycle scale are obtained by the multi-agent proximal policy optimization algorithm. Experimental results show that the proposed algorithm can improve the degree of service level agreement of service applications and the cluster resource utilization efficiency.