RSDD: Resilient Statistical Drift Detection for vSDN Hypervisors

Sixth-Generation (6G) virtualized Software-Defined Networks (vSDNs) require hypervisors that remain stable under non-stationary and bursty traffic. Existing systems, including DLVisor, still rely on a single Anderson-Darling test, which reacts poorly to gradual or variance-driven distributional shifts. This work introduces a unified drift-aware hypervisor, the Resilient Statistical Drift Detection (RSDD) framework, combining KSWIN, Page-Hinkley, Jensen-Shannon divergence, and variance monitoring into a consensus drift-severity index that guides an xLSTM forecaster and cost-aware slice reallocator. Experiments on Tier-2 ISP traces show RSDD significantly reduces overutilization events while maintaining SLA-compliant bandwidth allocation.