Improved Performance Control of Cloud-Native Microservices in the Edge with Proactive Autoscaling

Fodor, Balázs and Sonkoly, Balázs (2025) Improved Performance Control of Cloud-Native Microservices in the Edge with Proactive Autoscaling. In: NOMS 2025-2025 IEEE Network Operations and Management Symposium. IEEE IFIP Network Operations and Management Symposium . IEEE, Piscataway (NJ), p. 11073633. ISBN 9798331531638; 9798331531645

Text noms25-scaling.pdf Restricted to Repository staff only Download (1MB) | Request a copy

Abstract

Shifting from cloud to edge computing offers the advantage of being closer to the user, which improves latency and helps meet performance-related Service Level Agreement (SLA) requirements. However, the limited resources at the edge necessitate efficient resource scaling to handle fluctuating user demand. To maximize resource utilization, microservices architecture is favored over traditional monolithic approaches, allowing independent scaling of components so that only those needing extra resources are adjusted. Yet, standard reactive scaling methods may struggle to cope with unpredictable user traffic, leading to potential SLA violations. This underscores the need for proactive scaling solutions, where machine learning can play a key role in meeting diverse SLA requirements. In this work, we address these challenges by introducing a machine learning (ML) based proactive scaling framework for microservices in the edge. Our contribution is threefold, first we analyze several ML algorithms, identifying those that can be effectively applied in scaling. Second, we design and implement a scaling system that is capable of collecting metrics at multiple levels and making scaling decisions using ML models to ensure that the application meets the requirements specified in the SLA. Third, the system's efficiency is analyzed by measurements executed in a real environment, where we scale our test microservices-based application. Results show that the system can outperform the Kubernetes' Horizontal Pod Autoscaler in terms of SLA awareness without significant additional resource allocation, making it suitable for the edge.

Actions (login required)

Edit Item