Enhancing Kubernetes Traffic Distribution with a Dynamic Load Balancer Using Serverless Computing

Chatterjee, Ritika (2025) Enhancing Kubernetes Traffic Distribution with a Dynamic Load Balancer Using Serverless Computing. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration Manual) Download (828kB) | Preview

Abstract

Traditional Kubernetes LoadBalancer Services use round-robin or session-based algorithms for traffic distribution. While effective for stateless applications, this approach ignores real-time variations in pod resource utilization, allowing bottlenecks to emerge when some pods become overloaded and others remain idle. To address this critical gap, this paper introduces a dynamic metrics-driven load balancer. We integrate a Google Cloud Function as a serverless ingress point that delegates routing to a custom Python service within the Kubernetes cluster. This balancer queries a Prometheus monitoring stack for live pod CPU utilization and intelligently forwards each request to the least loaded pod, ensuring fair distribution and maximum resource efficiency. We validated the design on a three-node Google Kubernetes Engine cluster hosting an Nginx application and conducted controlled load tests comparing the default Kubernetes Service against our dynamic model. The dynamic architecture achieved 100% request success and maintained pod CPU variance within ±10%, whereas round-robin suffered a 20% failure rate and ±25% CPU variance. Conceptually, our work extends state-of-the-art by demonstrating a lightweight, non-intrusive mechanism for real-time traffic steering without ML (Machine Learning)/RL (Reinforcement Learning) overhead. Practically, it empowers operators to guarantee reliability and efficient utilization under heterogeneous workloads. Future research will explore multi-metric scoring (e.g., memory, network I/O) and adaptive caching to further reduce latency while preserving balanced distribution.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Pulido Gaytan, Luis Bernardo UNSPECIFIED
Subjects:	Q Science > QA Mathematics > Algebra > Algorithms 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:	20 Mar 2026 12:06
Last Modified:	20 Mar 2026 12:06
URI:	https://norma.ncirl.ie/id/eprint/9203

Actions (login required)

View Item