UURA: Uncertainty-Averse Unified Resource Allocation in Overcommited Private Clouds

This paper proposes UURA, an Uncertainty-averse Unified Resource Allocation algorithm designed for online services in overcommitted private clouds. Our algorithm incorporates pessimistic assumptions about node load uncertainty while evaluating CPU, memory, I/O resources and load-node fitness simultaneously. Experiment results on Kubernetes show that our algorithm achieve a more balanced load distribution across cluster without compromising the sub-second scheduling decision times.