Handling Real-time Slice requests with QoS-aware Offline Optimization and Online Adaptation

Abstract

Network slicing in sixth-generation (6G) systems requires dynamic and priority-aware resource reconfiguration to satisfy service-level agreements (SLAs) under varying traffic conditions. Existing optimization- and learning-based approaches often overlook scenarios where new slice requests cannot be accommodated by residual capacity, requiring controlled reallocation from existing slices. This paper develops a mathematical framework for slice reconfiguration that jointly allocates bandwidth and processing resources under end-to-end (E2E) delay and throughput constraints. Two offline formulations are presented: an auxiliary-variable model for exact constraint handling and a Softplus-based approximation that enables smooth gradient optimization. Building on the same structure, an online projected-gradient method is introduced to adapt allocations when new slices arrive, ensuring priority-aware and graceful quality-of-service (QoS) degradation under fixed resource budgets. Simulation results show that the offline and online methods exhibit consistent allocation behavior, with the online approach enabling near real-time reconfiguration while the offline formulations provide baseline references under complete demand knowledge. © 2025 IEEE.