HORIZON: Holistic Online Heuristic for Power and QoS-Aware Service Provisioning in Sliced 6G Networks

The Sixth Generation (6G) networks are envisioned to support diverse and stringent Quality of Service (QoS) demands through advanced technologies such as network slicing and Network Function Virtualization (NFV). However, the online provisioning of Service Function Chains (SFCs) in such dynamic, multi-domain environments presents a complex optimization challenge. Existing approaches often fail to holistically balance power consumption across both computational and networking domains while adhering to strict, per-request QoS guarantees and slice-aware resource sharing policies. This paper addresses this gap by proposing HORIZON, a novel and holistic online heuristic for power and QoS-aware SFC Embedding (SFCE) in sliced 6G networks. The primary objective is to jointly minimize the total incremental network power consumption, encompassing both servers and Reconfigurable Optical Add-Drop Multiplexers (ROADMs), and the service blocking rate. The problem is first formulated as an Integer Linear Program (ILP), which includes a detailed linearization of the non-linear ROADM power model, to serve as an optimal benchmark. The core contribution, HORIZON, employs a proactive, backward placement strategy guided by a multi-metric server ranking function and a segmental, backward, QoS-aware routing algorithm. Extensive discrete-event simulations across various network topologies and dynamic traffic loads validate the proposed approach. Results demonstrate that HORIZON significantly outperforms state-of-the-art benchmark heuristics (ONE and Holu), consistently achieving performance closest to the optimal ILP benchmark. In resource-constrained, large-scale scenarios, HORIZON achieves relative power savings of up to 23.6% compared to Holu and 12.8% compared to ONE, while maintaining the lowest and most stable service blocking rates. Furthermore, HORIZON proves to be highly robust and computationally efficient, processing requests up to 67x faster than the ILP benchmark, establishing it as a practical solution for real-time service provisioning in 6G systems.