eduzhai > Applied Sciences > Engineering >

Dynamic Graph Neural Network for Traffic Forecasting in Wide Area Networks

  • king
  • (0) Download
  • 20210507
  • Save

... pages left unread,continue reading

Document pages: 10 pages

Abstract: Wide area networking infrastructures (WANs), particularly science andresearch WANs, are the backbone for moving large volumes of scientific databetween experimental facilities and data centers. With demands growing atexponential rates, these networks are struggling to cope with large datavolumes, real-time responses, and overall network performance. Networkoperators are increasingly looking for innovative ways to manage the limitedunderlying network resources. Forecasting network traffic is a criticalcapability for proactive resource management, congestion mitigation, anddedicated transfer provisioning. To this end, we propose a nonautoregressivegraph-based neural network for multistep network traffic forecasting.Specifically, we develop a dynamic variant of diffusion convolutional recurrentneural networks to forecast traffic in research WANs. We evaluate the efficacyof our approach on real traffic from ESnet, the U.S. Department of Energy sdedicated science network. Our results show that compared to classicalforecasting methods, our approach explicitly learns the dynamic nature ofspatiotemporal traffic patterns, showing significant improvements inforecasting accuracy. Our technique can surpass existing statistical and deeplearning approaches by achieving approximately 20 mean absolute percentageerror for multiple hours of forecasts despite dynamic network traffic settings.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×