Multi-tenancy in Kubernetes for SaaS applications : deployment, scaling, and resource optimization using Helm and namespaces

Abstract

This thesis explores the implementation and evaluation of a multi-tenant Software as a Service platform built on Kubernetes using namespace-based isolation mechanisms. The research aims to fill a gap in existing literature by providing empirical validation of namespace isolation's effectiveness under severe resource constraints, specifically using Amazon Web Services Free Tier infrastructure with a single-node k3s cluster. The study uses a method that is focused on engineering and combines designing and implementation of the system. A platform with 3 parts was made using Node.js and Express. This platform was then packaged up using Docker. Comprehensive testing was conducted over a 204 minutes observation period, evaluating performance characteristics, autoscaling behavior, resource fairness, security isolation, and system reliability. The platform achieved 891 requests per second on the API service and 412 requests per second on the authentication service, with 95th percentile latencies of 135 milliseconds and 279 milliseconds respectively. The results demonstrate that namespace-based multi-tenancy provides capabilities for small to medium-scale SaaS platforms serving trusted tenants. The findings challenge common assumptions that physical isolation is universally necessary, proving that logical isolation mechanisms achieve sufficient security and performance when properly configured. This thesis contributes a replicable architectural trace, complete implementation artifacts, and comparative data.