Dynamic Multi-Provider Cluster Autoscaling For The Computing Continuum

Although existing Kubernetes cluster autoscalers dynamically adjust resources to match demand, they remain confined to the local provider's administrative domain, leading to inefficiencies, vendor lock-in, and are not appropriate for edge clusters. This paper introduces a novel multi-provider autoscaling architecture that enables the existing autoscaling paradigm to operate across multiple, independent federated clusters. The proposed solution leverages Liqo, an open-source framework enabling seamless cross-cluster federation, to expose remote compute capacity as virtual nodes within the local control plane, hence achieving transparent resource sharing. We design and implement a Liqo-aware Cluster Autoscaler that integrates remote resource discovery, enables policy-driven scaling decisions, and cost-aware placement across heterogeneous environments. Preliminary experimental evaluation demonstrates that our approach drastically improves resource utilization, especially in highly dynamic environments.