Performance Analysis of High Availability Video Conference Service with Kubernetes Cluster Across Data Center

Abstract

Due to increasing number of users and network traffics, video conference services can experience downtime in case of overloading request to video conference servers. To avoid this, an infrastructure that can replicate itself is for mitigating the risk of service downtime. The infrastructure is built using a container orchestration tool called Kubernetes that runs on Digital Ocean and Linode which functions as Cloud Services. This paper discusses the design of a Kubernetes cluster in which there are two Data Centers originating from two different countries. The Data Center can accommodate several Kubernetes nodes where each node accommodates one pod that can run one application through an intermediary service. In addition, nodes are equipped with the auto-scaling feature, i.e., ability to replicate pods and resize computational resources automatically, which can increase the high availability of the infrastructure and mitigate the risk of downtime. As for the experiment, The WebRTC service is observed by adjusting several parameters, which are QoS (delay, jitter, packet loss), CPU and memory usages, and the number of pods. The experimental results show that the number of pods before the existence of HPA was one while after the existence of HPA there were four pods. The highest number of errors before the HPA was 490 responses/sec, while after the HPA there were 11 responses/sec.