Kubespace Achieves 59% Latency Reduction for LEO Satellite Container Orchestration

Researchers are tackling the challenges of utilising low Earth orbit (LEO) satellites for cloud computing, a field poised to deliver global connectivity and support for the Internet of Things. Zhiyuan Zhao, Jiasheng Wu, and Shaojie Su, alongside Wenjun Zhu et al from the Institute of Space Internet and College of Computer Science and Artificial Intelligence at Fudan University, China, present KubeSpace, a novel control plane designed to overcome the high latency and instability issues inherent in orchestrating containers across these geographically dispersed and frequently changing LEO networks. This work is significant because KubeSpace demonstrably improves management latency by 59% and eliminates management interruption, paving the way for reliable and efficient space cloud infrastructure.

The research team tackled this issue by proposing KubeSpace, a system specifically engineered for the unique demands of LEO satellite container orchestration. This contrasts with single control node systems where communication can require numerous inter-satellite hops, introducing significant latency. Furthermore, the orbit-aware placement strategy proactively minimises both communication latency and the frequency of handovers by intelligently assigning optimal controllers based on predictable satellite trajectories.

This proactive approach anticipates and mitigates potential disruptions, enhancing the stability of the control plane. Extensive experiments utilising real satellite traces demonstrate the effectiveness of KubeSpace compared to existing solutions. The study reveals a substantial 59% reduction in the average management latency of satellite nodes, a critical improvement for responsive onboard application management. Crucially, KubeSpace achieves this latency reduction without experiencing any management interruption time, a significant advancement over traditional multi-controller architectures that suffer from temporary loss of satellite visibility during handovers.

This seamless handover capability is achieved through careful design of the control node assignment and transition processes. The architecture of KubeSpace centres around optimising the multi-control node approach, enabling smooth transitions between controllers as satellites move within the LEO constellation. The team implemented a prototype based on Kubernetes v1.31.10, validating their approach with realistic data. The research establishes that KubeSpace not only reduces latency but also decreases control node handover time by 84%, ensuring continuous operation of onboard containers and resources. This breakthrough opens possibilities for more complex and reliable space-based applications, including AI inference and lightweight 5G networks, all managed efficiently from the ground.

Distributed ground control for LEO container orchestration enables

This approach contrasts with single control node systems, which suffer from high management latency due to potential inter-satellite hops requiring several hundred milliseconds. To overcome this, the study pioneered a multi-control node deployment, strategically positioning ground stations to minimise communication delays. However, traditional multi-controller architectures present challenges during satellite handovers, causing temporary loss of control plane visibility and disruption to onboard applications. KubeSpace overcomes this by enabling seamless handover between control nodes, preventing management interruptions during transitions.

The team designed a workflow where, upon handover decision, the system initiates pod scheduling and enables watch events on the target control node, concurrently draining and evicting pods from the source node. This process, detailed in Figure 3a, includes cleaning up resources and establishing communication channels, minimising downtime. Experiments revealed a time overhead breakdown, with pod restart taking 1.12 seconds, draining and eviction 1.65 seconds, cleanup 5.70 seconds, and joining the running node completed in 9.70 seconds. This minimises both communication latency and the frequency of handovers, enhancing overall system stability. The KubeSpace prototype was built upon Kubernetes v1.31.10 and rigorously tested using real LEO satellite constellation data. Crucially, the research achieved complete elimination of management interruption time during satellite handovers, enabling stable control over onboard resources and containers.

KubeSpace cuts LEO satellite management latency significantly

This breakthrough delivers a significantly more stable and responsive system for managing satellite-based cloud computing resources. The team measured management latency and handover times using real satellite trace data to validate the performance of KubeSpace. Results demonstrate an 84% decrease in control node handover time, a critical improvement for maintaining continuous operation during satellite transitions between ground stations. This innovative approach avoids the visibility loss and application disruption common in traditional multi-controller setups, where satellites must disconnect from one cluster before joining another.

Further tests confirm that KubeSpace completely eliminates management interruptions during satellite handovers, a substantial advancement over conventional methods. The system’s orbit-aware placement with a dynamic assignment strategy minimizes both latency and handover frequency by leveraging predictable satellite trajectories to assign optimal controllers. Data shows that this proactive assignment significantly reduces communication delays, which can reach several hundred milliseconds in systems relying on numerous inter-satellite hops. The prototype implementation, based on Kubernetes v1.31.10, was rigorously tested with real LEO satellite constellation data to ensure its effectiveness.

Scientists designed a workflow for control node handover, detailing the steps from initiating the handover to establishing communication channels with the target control node. Measurements of the time overhead associated with each step, restarting pods, draining and evicting containers, cleaning up resources, and joining the new node, were recorded and analyzed. The research highlights that traditional platforms experience significant delays during these processes, leading to periods where the pod is unavailable and the node is invisible to the control plane, whereas KubeSpace mitigates these issues. This work establishes a foundation for reliable and efficient space cloud computing, enabling advanced applications and services in LEO.

LEO Constellation Orchestration via KubeSpace Design simplifies application

The authors acknowledge limitations related to the scope of their experiments, focusing on specific trace data and network configurations. Future work could explore the system’s performance with a wider range of applications and more complex orbital scenarios, as well as investigating integration with diverse satellite communication protocols. These findings represent a step towards realising the potential of space cloud computing by providing a stable and efficient platform for managing containerised applications in LEO.