Reinforcement Learning-based Adaptive Path Selection for Programmable Networks

2025-08-19

AI Summary

To address path-selection lag under dynamic congestion and high overhead from centralized control in programmable networks, this paper proposes a distributed adaptive routing framework that synergistically integrates Stochastic Learning Automata (SLA) with In-band Network Telemetry (INT). Implemented on P4-programmable BMv2 switches, the framework leverages INT to collect fine-grained, real-time link-state telemetry directly in the data plane, enabling lightweight online learning and forwarding decisions by SLA locally—without controller involvement. Its key innovation lies in the first deep integration of SLA and INT, achieving congestion-aware path adaptation at line rate with low overhead, full distribution, and rapid convergence (millisecond-scale). Experimental evaluation demonstrates that, compared to static routing and ECMP, the approach improves network throughput by up to 37% and reduces average flow completion time by 29%, significantly enhancing resource utilization and robustness under dynamic traffic conditions.

Technology Category

• Planning, Routing, and Scheduling: Planning/Scheduling and Learning
• Machine Learning: Online Learning & Bandits
• Intelligent Robots: Learning & Optimization for ROB

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems

Abstract

This work presents a proof-of-concept implementation of a distributed, in-network reinforcement learning (IN-RL) framework for adaptive path selection in programmable networks. By combining Stochastic Learning Automata (SLA) with real-time telemetry data collected via In-Band Network Telemetry (INT), the proposed system enables local, data-driven forwarding decisions that adapt dynamically to congestion conditions. The system is evaluated on a Mininet-based testbed using P4-programmable BMv2 switches, demonstrating how our SLA-based mechanism converges to effective path selections and adapts to shifting network conditions at line rate.

Problem

• Adaptive path selection in programmable networks
• Dynamic congestion adaptation using reinforcement learning
• Real-time telemetry-driven forwarding decisions optimization

Innovation

• Reinforcement learning for adaptive path selection
• Combines learning automata with real-time telemetry
• P4-programmable switches enable line-rate decisions