BeNNS: A Surrogate Model for Hybrid Online-Offline Evolution of SFC Embedding

2025-09-20

AI Summary

SFC embedding into physical networks is NP-hard, and conventional genetic algorithms (GAs) suffer from prohibitive online fitness evaluation overhead, hindering real-time deployment. This paper proposes a hybrid online-offline evolutionary framework, whose core innovation is a topology- and traffic-agnostic BeNNS surrogate model: a neural network trained offline to approximate the fitness function, enabling rapid online solution quality assessment. To our knowledge, this is the first GA-based approach applicable to online SFC embedding under dynamic network conditions. Experiments demonstrate that our method identifies deployable configurations among thousands of candidates in only 36.8 minutes on average—nearly 30× faster than pure online GA (17.9 hours), which also exhibits poor convergence. The framework thus significantly enhances both timeliness and practicality for real-world SFC orchestration.

Technology Category

• Search and Optimization: Evolutionary Computation
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

Service Function Chains (SFCs) enable programmatic control of the functions and services in a computer network. By leveraging Software Defined Networking to control the links between virtualised network functions, SFCs provide a scalable approach to dealing with the increased pressures on network operation and management. Unfortunately, the challenge of embedding SFCs onto the underlying physical network and compute infrastructure is an NP-hard problem. Genetic Algorithms (GAs) have been used to address this issue, but they require significant time to evaluate solution quality (fitness) extit{online}, with most solutions instead adopting extit{offline} simulations or analytical evaluations. To enable online use of GAs in solving the SFC embedding problem, we introduce a hybrid online-offline approach to evaluate generated solutions. At the core of this is BeNNS--a topology, traffic, and SFC-embedding agnostic surrogate model that approximates fitness. We evaluate our approach across six experiments, varying available resources and traffic loads. Our results demonstrate that our approach is capable of exploring thousands of potential configurations and generating deployable solutions in 36.8 minutes on average, compared to online-only approaches, which take 17.9 hours on average to explore tens of solutions, which do not converge on an optimal solution.

Problem

• Addressing NP-hard SFC embedding onto physical network infrastructure
• Enabling online genetic algorithms for real-time SFC configuration optimization
• Developing surrogate model to approximate fitness evaluation for faster convergence

Innovation

• Hybrid online-offline evolution for SFC embedding
• BeNNS surrogate model approximates fitness agnostically
• Enables exploring thousands of configurations in minutes