In-Network Artificial Computing Enhanced Light Model-Switching for Emergency Communications Networks

2026-05-11

AI Summary

This work addresses the sluggish response of conventional model-update mechanisms in emergency communication networks, which struggle to meet real-time intelligent processing demands under dynamic traffic conditions. The authors propose an in-network AI computing framework that enables lightweight model switching, achieving for the first time microsecond-scale, interruption-free online switching among multiple binary neural networks on general-purpose hardware. Leveraging eBPF/XDP and AF_XDP, along with Binary Neural Network design, x86 AVX-512 instruction set optimizations, and fixed 1024-byte payload alignment, the framework incurs a switching overhead of merely 0.005 µs with zero misclassified packets. Even with 16 models coexisting, it sustains high efficiency—achieving a throughput of 1.894 Mpps and an inference latency as low as 0.528 µs—thereby significantly enhancing real-time responsiveness in emergency networks.

Abstract

Emergency communications networks require in-network intelligence for timely traffic handling under dynamic demands and runtime constraints. In these environments, packets may need different inference behaviors, and conventional model replacement via control-plane updates is too slow for responsive operation. We propose an in-network artificial computing framework with lightweight model-switching, where multiple Binary Neural Network (BNN) models are kept resident within a shared execution framework. Packet metadata selects the active model at packet granularity with O(1) selection cost. A fixed 1024-byte payload is aligned with x86 AVX-512, enabling efficient memory access. The framework is realized on an eBPF/XDP + AF_XDP stack. Experimental results show that the system sustains 1.894 Mpps with a 0.528 us inference latency, while model selection adds only 0.005 us. Our results demonstrate that different resident models induce distinct packet-processing behaviors, that scaling to 16 slots preserves low switching overhead, and that online model switching completes without wrong-verdict packets. These results show the practicality of lightweight in-network artificial computing on commodity hardware.

Problem

• Emergency communications networks
• In-network intelligence
• Model-switching
• Real-time inference
• Dynamic traffic handling

Innovation

• in-network computing
• model-switching
• Binary Neural Network
• eBPF/XDP
• emergency communications