Analyzing QoS Performance in Kubernetes-Based High Scalability Clusters

Abstrak

This paper proposes a performance analysis of the auotscaling methods that can be utilized in Kubernetes Clusters in order to determine cost-efficiency for services. This paper aims to accommodate the specifications for an analysis by using Google Kubernetes Engine (GKE) built into Google Cloud Platform (GCP) as a platform to orchestrate clusters alongside deployed services. To fulfill the cluster model, customized cluster settings and NGINX-based services are developed to manage the website used for load testing in order to determine the behavior of the autoscaling methods. This paper proposes an experiment using real-time transmission of Virtual Users (VU) sent from an application, called k6, used to simulate a number of users accessing the website within a time frame through a public network. During the experiment, the quality of service of the website is measured and analyzed using metrics and events from GKE and the results made by k6 to determine whether the autoscaling system works within set parameters and determine cost efficiency in Kubernetes cluster. The proposed load test is able to ensure that the number of HTTP requests results in HTTP Status Code 200 and that 95% of the requests are completed in less than 2 seconds.