Scaling a Kubernetes Cluster

Kubernetes is a container orchestration tool that has become widely adopted for deploying and scaling containers. Devatus Oy as well as their subsidiary company Fliq Oy are interested in knowing how containerized applications can be scaled on Kubernetes. The objective of this the-sis is to research how a Kubernetes cluster can be scaled as well as containerized applications running on Kubernetes. This thesis begins with an introduction to necessary background knowledge needed to under-stand what Kubernetes is. Cloud computing and distributed systems are introduced, since Ku-bernetes is a distributed system used in cloud environments for the most part. Furthermore, distributed applications and workloads are introduced through the concept of microservices. The concept of containerizing applications is thoroughly introduced to understand the runtime environment of the applications deployed to Kubernetes. Finally, Kubernetes architecture as well as its main components are introduced to understand how container orchestration works. The research on Kubernetes scalability is divided into three different parts. First part consists of researching how containerized applications can be scaled on Kubernetes. Second part is focused on how the Kubernetes cluster itself can be scaled. The final part consists of load testing one of Fliq’s example REST API applications deployed to a local Kubernetes cluster. The purpose of load testing is to gain further insight into scaling applications running on Kubernetes. Load test results are compared between the initial deployment configurations and after scaling the application. The load test results show that containerized applications can be scaled both vertically and hor-izontally. Vertical scaling can be achieved by increasing the requested and limited CPU and RAM resources for a Pod. Horizontal scaling can be achieved by increasing Pod replicas as well as having a Service in front of the Pods that load balances the incoming traffic. Load test results show that both vertical and horizontal scaling can increase the number of users supported by an application deployed to Kubernetes. Scaling horizontally is preferred for Fliq’s example REST API since it decreased average response time and increased throughput