Agent-Based Multi-Objective Optimization for Network Slicing and Admission Control in 5G/6G Softwarized Infrastructures

Abstract

Fifth-generation and prospective sixth-generation mobile networks are based on highly softwarized infrastructures in which network functions and resources are dynamically instantiated and reconfigured across disaggregated computing and transport domains. Network slicing exposes these capabilities as logically isolated end-to-end tenants, each with distinct performance, reliability, and cost requirements. Admission control must decide whether new slice requests can be accepted without violating guarantees for existing slices, while resource allocation mechanisms must coordinate heterogeneous compute, storage, and radio resources. Multi-objective trade-offs arise naturally between spectral efficiency, latency, energy consumption, and operational expenditure. Centralized optimization approaches struggle with the dimensionality, dynamics, and partial observability of realistic 5G and 6G deployments. Agent-based coordination, where distributed software agents represent slices, infrastructure domains, and orchestration functions, provides a flexible means to adapt to local conditions, negotiate resource usage, and incorporate learning-based policies. This work examines an agent-based multi-objective optimization framework for network slicing and admission control in softwarized infrastructures. The study focuses on linear models of resource coupling and service-level constraints, explores scalarization and constraint-based formulations of multiple objectives, and considers both myopic and horizon-based decision processes. The discussion emphasizes how autonomous agents can approximate global optimization goals with limited information exchange and modest computational overhead, and how linear relaxations can support explainable admission decisions and capacity planning. The analysis covers system modeling, algorithmic design, and qualitative evaluation of trade-offs, aiming to clarify the conditions under which agent-based strategies achieve consistent and predictable behavior in the presence of rapidly varying traffic demands and slice heterogeneity.