Core Concepts
The proposed solution leverages Cluster API and Liqo to enable efficient multi-cluster management and interconnectivity, catering to the stringent requirements of distributed Extended Reality (XR) services.
Abstract
The article explores the challenges of supporting Extended Reality (XR) services, which have demanding Quality of Service (QoS) and functional requirements, in a multi-cluster deployment scenario. It identifies the limitations of contemporary Kubernetes in facilitating cross-cluster networking and management.
The key highlights are:
Examination of state-of-the-art multi-cluster management frameworks, such as KubeFed, Karmada, and Terraform, and their limitations in addressing the needs of XR services.
Exploration of multi-cluster interconnectivity solutions, including service mesh approaches (Istio, Linkerd, Consul) and overlay network solutions (Submariner, Skupper, Liqo).
Proposal of a novel solution combining Cluster API for multi-cluster orchestration and Liqo for cross-cluster networking, which addresses the specific requirements of XR services.
Experimental evaluation of the proposed solution in the context of a cross-cluster video streaming use case, assessing the provisioning times, end-to-end latency, and resource consumption.
The results demonstrate the efficiency of the proposed solution in supporting the QoS and functional requirements of distributed XR services, including low latency, UDP support, and dynamic, multi-ownership deployment scenarios.
Stats
For an end-user experience to be considered satisfactory, the end-to-end latency shall not be greater than 15ms, and the available bandwidth should be scalable up to 30 Gbps.
The 50th percentile of end-to-end latency for the cross-cluster video streaming use case ranged from 794ms to 824ms, with the 90th percentile consistently around 1 second.
The CPU usage for the cross-cluster scenarios had a 50th percentile between 29.4% and 33.5%, while the memory usage varied between 62% and 70%.
Quotes
"Kubernetes is extremely popular in cloud computing environments, lightweight versions, such as K3s, are often deployed in Edge computing environments."
"To ensure seamless cross-cluster communications in multi-cluster deployments, attention is needed for both cluster management and connectivity between clusters."
"Liqo manages to satisfy all of the functional requirements that are associated with XR services in terms of providing support for UDP, establishing a singular cross-cluster control plane for optimal workload scheduling, and facilitating dynamic, multi-ownership deployment scenarios."