Multi-objective Optimization for SFC Placement in Dynamic Cross-Domain Networks

Efficient service function chaining (SFC) placement is critical for optimizing network service delivery by enhancing the efficiency of data processing, particularly under block-oriented paradigms in the context of 5G and beyond networks. However, existing approaches face significant challenges in achieving effective multi-objective SFC placement within dynamic cross-domain networks (CDNs), such as the absence of robust mechanisms to accurately assess the follow-up computational resource availability under resource-constrained conditions, with the objective of latency optimization. To overcome these challenges, we propose a novel architectural framework designed to optimize SFC placement while considering both resource constraints and latency in dynamic CDNs. A dedicated functional block that integrates Transformer-based deep learning models with linear programming (LP), introducing a resource fragmentation factor, is incorporated into the framework to enable more informed placement decisions. Then, a dynamic modeling system is developed, leveraging domain-aware detection and domain topology encoding via graph autoencoders (GAEs), to facilitate real-time adaptation to evolving network topologies. After that, the Pareto frontier is employed to ensure optimal tradeoffs between resource utilization and latency. In each iteration of virtual network function (VNF) deployment, the domain detection module continuously updates environmental metrics while the topology encoder dynamically captures emergent domains. To further accelerate decisionmaking, a Transformer-based module is used to generate optimized placement solutions. Extensive experiments have been conducted across three physical network scales and benchmarked against three baseline schemes, demonstrating the superiority of the proposed framework in latency and deployability in dynamic CDNs.