A Design and Development of Operator for Logical Kubernetes Cluster over Distributed Clouds
Thanh-Nguyen Nguyen, Jangwon Lee, Vitumbiko Mafeni, Younghan Kim · 2024
The conventional method of managing networks and infrastructure involves multiple components that primarily include administration experts alongside automated scripts and functions to manage, track, monitor, and resolve incidents and fine-tune configurations of network and computing resources. Several platforms and research work in the field of multi-cluster management have surfaced; however, these often necessitate pre-existing Kubernetes clusters, involve intricate configurations, exhibit inflexibility, and lack robust automation for cluster provisioning and management. Moreover, these platforms lack a networking solution for inter-cluster connectivity or do not fully manage the lifecycle of the Kubernetes cluster. Consequently, we introduce Logical Kubernetes as a Service (L-KaaS), an autonomous orchestration platform for managing distributed Kubernetes clusters. This platform features a ’Logical Kubernetes Cluster’ - a declarative specification for managing a huge number of distributed Kubernetes clusters with the same simplicity as managing a single Kubernetes cluster that spans across all locations. We developed operators behind Logical Cluster to provide a simplified, abstract interface via a declarative application programming interface (API) to automate software configuration and workload orchestration tasks and absorb dynamic infrastructure variations. Regarding inter-cluster connectivity, we used a high-performance, scalable, and consistent networking service mesh solution that relies on new kernel technology -eBPF. In workload scheduling tests, our system outperformed existing solutions, showcasing a performance that was 8.86 times faster than Kubernetes and 1.27 times faster than the Oaskestra scheduler.