A Latency‐Prediction‐Based Speculative Multipath Routing Method for SDN

Abstract

This paper proposes Speculative Multipath Routing based on Latency Prediction (SMR‐LP), a novel routing method designed for Software‐Defined Networking (SDN) environments. Traditional SDN routing strategies typically rely on static path selection based on historical traffic statistics, which restricts their adaptability to dynamic network conditions and failure scenarios. In contrast, SMR‐LP integrates latency and packet loss measurements with time‐series forecasting to enable proactive and adaptive path selection. Leveraging Adaptive Exponential Smoothing Moving Average (AESMA) model, SMR‐LP predicts future network states and evaluates candidate paths using a weighted cost function that incorporates both latency and reliability. Path selection is dynamically executed through OpenFlow's SELECT group mechanism, providing efficient traffic splitting and failover support. To further reduce computational complexity in path exploration, we propose Fast Branch Hop based Depth‐First Search (FBDFS) algorithm, which applies pruning techniques to efficiently compute cost‐optimal paths for each first‐hop branch. Emulation‐based experiments under realistic traffic scenarios, including link failures and recovery, demonstrate that SMR‐LP outperforms conventional Shortest Path First (SPF) and multipath routing (MPR) in terms of throughput and packet delivery ratio. Experimental results highlight SMR‐LP's predictive capability, robustness, and adaptability, making it a promising routing strategy for latency‐sensitive and mission‐critical SDN applications. © 2026 Institute of Electrical Engineers of Japan and Wiley Periodicals LLC.