Machine Learning – Driven Auto Scaling in Kubernetes for Cloud Resource Optimization

Shroff, Ankur (2025) Machine Learning – Driven Auto Scaling in Kubernetes for Cloud Resource Optimization. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (2MB) | Preview
Preview	PDF (Configuration Manual) Download (12MB) | Preview

Abstract

In response to high fluctuation of user-traffic, cloud-computing services face the biggest challenge in managing dynamic resource allocation in Kubernetes as it uses default Horizontal Pod Auto-Scale (HPA) which only reacts to scale up or down depending on performance metrics based on limit threshold reached of traffic-load utilizing CPU, which leads to service degradation along with poor user experiences. In this research, auto scaling system design is more proactive and hybrid to overcome these limitations. This designed system that uses time-series forecasting model - Prophet, LSTM, ARIMA, which are trained on historical real-world cloud usage data – Alibaba 2018 Cluster and synthetic traffic-loads values, to predict future demand of system scaling. Thus, prediction is used to scale pods, where reactive component acts as a safety of scaling-down after traffic-load reduces. Performance of system was evaluated in Kubernetes environment based on 3 different phases from standard auto-scaler to ML-predictions. To replicate the real-world traffic-load, tools are used to run the traffic-load on system to test the reliability of Kubernetes system based on LSTM-model with future demands of 10-second. Proactive scaler does improve resources allocation, resource utilization, anticipating traffic-load spikes and reducing the latency.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Siddig, Abubakr UNSPECIFIED
Uncontrolled Keywords:	Kubernetes(K8); Docker; Hey; Prometheus; Grafana; HEY; Push Gateway; Python; Django
Subjects:	T Technology > T Technology (General) > Information Technology > Cloud computing Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Master of Science in Cloud Computing
Depositing User:	Ciara O'Brien
Date Deposited:	01 Sep 2026 11:04
Last Modified:	01 Sep 2026 11:04
URI:	https://norma.ncirl.ie/id/eprint/9737

Actions (login required)

View Item