Service Function Chaining Architecture for Multi-hop Split Inference and Learning

Abstract

Service Function Chaining (SFC) is a networking technique that ensures traffic traverses a predefined sequence of service functions, realizing arbitrary network services through dynamic and efficient communication paths. Inspired by this concept, we propose an SFC-based architecture for Multi-hop Split Inference (MSI), where split sub-models are interpreted as service functions and their composition forms a service chain representing the global model. By leveraging SFC, the proposed architecture dynamically establishes communication paths for split sub-models, ensuring efficient and adaptive execution. Furthermore, we extend this architecture to Multi-hop Split Learning (MSL) by applying SFC to the bidirectional communication required for training tasks. To realize the proposed architecture, we design Neural Service Functions (NSFs) to execute split sub-models as transparent TCP proxies and integrate them with Segment Routing over IPv6 (SRv6) and the extended Berkeley Packet Filter (eBPF)-based SFC proxy. This integration ensures efficient ML processing over dynamic routing while maintaining compatibility with existing applications. Evaluation results demonstrate that (1) the proposed architecture is feasible for both MSI and MSL; (2) it is particularly suitable for real-time inference in MSI scenarios with small mini-batch sizes; (3) it supports dynamic path reconfiguration, enabling adaptive responses to changing network conditions while minimizing the impact of control mechanisms on inference and learning processes.