Towards resource-efficient reactive and proactive auto-scaling for microservice architectures
Hussain Ahmad, Christoph Treude, Markus Wagner, Claudia Szabo
Abstract
Microservice architectures have become increasingly popular in both academia and industry, providing enhanced agility, elasticity, and maintainability in software development and deployment. To simplify scaling operations in microservice architectures, container orchestration platforms such as Kubernetes feature Horizontal Pod Auto-scalers (HPAs) designed to adjust the resources of microservices to accommodate fluctuating workloads. However, existing HPAs are not suitable for resource-constrained environments, as they make scaling decisions based on the individual resource capacities of microservices, leading to service unavailability, resource mismanagement, and financial losses. Furthermore, the inherent delay in initializing and terminating microservice pods hinders HPAs from timely responding to workload fluctuations, further exacerbating these issues. To address these concerns, we propose Smart HPA and ProSmart HPA, reactive and proactive resource-efficient horizontal pod auto-scalers respectively. Smart HPA employs a reactive scaling policy that facilitates resource exchange among microservices, optimizing auto-scaling in resource-constrained environments. For ProSmart HPA, we develop a machine-learning-driven resource-efficient scaling policy that proactively manages resource demands to address delays caused by microservice pod startup and termination, while enabling preemptive resource sharing in resource-constrained environments. Our experimental results show that Smart HPA outperforms the Kubernetes baseline HPA, while ProSmart HPA exceeds both Smart HPA and Kubernetes HPA by reducing resource overutilization, overprovisioning, and underprovisioning, and increasing resource allocation to microservice applications. • Hierarchical architecture based horizontal pod auto-scaler. • Reactive and proactive resource-efficient auto-scaling policies. • Reactive auto-scaler outperforms Kubernetes baseline auto-scaler. • Proactive auto-scaler outperforms both reactive and Kubernetes auto-scalers. • Reduction in resource overutilization, underprovisioning, and overprovisioning.