Toward Highly Scalable Load Balancing in Kubernetes Clusters
Nguyen Nguyen, Taehong Kim
Abstract
Container-based virtualization has recently been flourishing due to its ease of deployment and flexibility of resource provisioning. Kubernetes, a well-known open source project, is an orchestration platform for containerized applications. Herein, applications can have multiple replicas to provide high scalability and availability, but some applications for stateful service require a leader to be elected to maintain consistency and coordinate tasks among the replicas. In this article, the Kubernetes architecture is first analyzed by focusing on load balancing, a leader election algorithm, and a leader-based consistency maintenance mechanism. Then we address the following challenges: the leader has heavy loads due to its inherent design, and the Kubernetes leader election algorithm cannot evenly distribute the leader throughout nodes. Finally, experimental results are provided to prove the importance of leader distribution throughout nodes in the cluster for highly scalable load balancing.