.png){kind=icon}
Return
Proactive SDN traffic management for IoT using temporal deep learning
.png){kind=icon}
OA
AI
DOI：10.1016/j.array.2026.101223.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Software-defined networking (SDN) provides centralized control for IoT traffic management; however, its conventional reactive scheduling behavior is often insufficient for time-varying, bursty, and latency-sensitive IoT traffic. The central purpose of this study is to develop a proactive SDN-based scheduling framework that can anticipate traffic changes in advance and translate these predictions into timely installations of OpenFlow rules. To address this objective, this paper proposes ATSDSO, an adaptive temporal deep scheduling optimizer that integrates temporal traffic prediction with control-aware scheduling for SDN-enabled IoT networks. The main innovation of ATSDSO lies in its closed-loop integration of four adaptive functional components: the Temporal Sensitivity Layer (TSL), Stream Variance Regularization (SVR), Adaptive Learning Rate Controller (ALRC), and Progressive Scheduling Reinforcement (PSR). Unlike existing DNN/LSTM-based or reactive SDN scheduling methods that usually treat prediction and control as separate processes. The proposed algorithm is trained and evaluated using normalized and standardized IoT traffic traces and is further tested under multiple OpenFlow timeout configurations to examine its usefulness under complex working conditions, including dynamic traffic variation, burst-oriented stream behavior, and different proactive scheduling intervals. Experimental results show improvement in traffic-intensity classification, inter-arrival time prediction, and latency-aware scheduling performance. The proposed framework achieves 97% classification accuracy, improves response time by 31%, reduces MAE by 44.6%, and reduces RMSE by 40.0% compared with existing DNN/LSTM-based scheduling approaches. In comparison with reactive SDN-based models, ATSDSO reduces RTT by 46.9%, 53.7%, and 60.6% under 30 s, 60 s, and 120 s OpenFlow timeout configurations, respectively. These improvements demonstrate that proactive rule installation can reduce controller delays and improve latency-aware scheduling. In addition, the adaptive functional component analysis confirms that TSL, SVR, ALRC, and PSR evolve in a stable and coordinated manner during training, supporting the robustness of proposed framework for temporal prediction and proactive SDN based IoT traffic management.
Keywords:
Software-defined networking
IoT traffic
Temporal traffic prediction
Proactive rule scheduling
Scheduling metrics
Journal
IF:
4.5
Papers:
934
Citations:
1.2K
Organization
No organization information available
Cited Papers
No cited papers available