A novel multi-authority access control scheme for fine grained access to users data in the cloud-based storage
TL;DR
Results demonstrate that HPA effectively responds to workload increases by provisioning additional pods, maintaining system stability and throughput during high-demand periods, with CPU usage and RPS exhibiting predictable scaling behavior aligned with the 15-second Metrics Server scraping interval.
Abstract
The widespread adoption of microservices architectures on Kubernetes has introduced significant challenges in resource management, particularly the inadequacy of default load balancing under dynamic workloads and issues with pod resource sharing under contention. This paper proposes an integrated auto-scaling platform that combines the Horizontal Pod Autoscaler (HPA), Metrics Server, and Prometheus to dynamically optimize resource utilization in a Kubernetes-in-Docker (KIND) cluster. The experimental platform was deployed on an Ubuntu 22.04 host with a three-node KIND cluster (one master, two workers), using Kubernetes v1.27.3 and Docker v24.0.7, with performance evaluated through CPU utilization and requests per second (RPS) metrics collected via Prometheus and visualized in Grafana. Results demonstrate that HPA effectively responds to workload increases by provisioning additional pods, maintaining system stability and throughput during high-demand periods, with CPU usage and RPS exhibiting predictable scaling behavior aligned with the 15-second Metrics Server scraping interval. The novelty of this work lies in the systematic integration of HPA with Prometheus custom metrics within a KIND environment, extending evaluation across multiple lightweight Kubernetes distributions including microk8s and minikube. This approach enhances scalability, computational efficiency, and cost-effectiveness for microservice-based systems.