Microservices Scheduling with ReS Model: Characterization, Regulation and Metrics

Abstract

Microservice architecture has become the dominant paradigm for cloud-native applications, yet is often subject to resource fragmentation when Horizontal Pod Autoscaler (HPA) is used. While Vertical Pod Autoscaler (VPA) addresses this by adjusting pod resources, the Re-Create approach causes downtime, impacting availability. In contrast, In-Place VPA allows for faster scaling without downtime but introduces the risk of redundant pod creation due to scaling-up failures. These redundant pods increase memory consumption, especially in production scenarios. This paper proposes the Resource Stage (ReS) Model to address this challenge. The ReS Model characterizes microservices’ resource usage patterns, enabling tailored resource regulation strategies to minimize redundant pod creation. Additionally, we introduce the Node-level ReS Distribution quantile as the node stress indicator for future resource demand statistics. Our experiments show that KubeReS, based on the ReS Model, enhances resource utilization by 40.9% compared to Kubernetes HPA and reduces per-node peak memory requirement by 34.7% – 43.7% compared to Hyscale.