BOW: First Real-World Demonstration of a Bayesian Optimization System for Wavelength Reconfiguration
Overview
We demonstrate a practical Bayesian Optimization system for wavelength reconfiguration at Facebook backbone. Our system uses a firewall for safe deployment. It is open-source, compatible with any vendor, and achieves 4.76x faster wavelength reconfiguration.
Diagram
Insights
  • Today's wavelength provisioning is slow (tens of minutes), mainly because of amplifier reconfigurations.
  • Today's optical line systems rely on proprietary amplifier control method based on complex modeling of amplifiers (noise figures, gain profiles)..
  • BOW leverages Bayesian Optimization for in-situ learning of amplifier control strategies.
  • BOW is production-ready and compatible with multi-vendor optical devices.
  • Artifact
    Your browser does not support SVG Code available: bow
    Publication
    BOW: First Real-World Demonstration of a Bayesian Optimization System for Wavelength Reconfiguration
    Z. Zhong, M. Ghobadi, M. Balandat, S. Katti, A. Kazerouni, J. Leach, M. McKillop, Y. Zhang
    OFC 2021
    Postdeadline Paper
    Paper | Slides | Artifact | BibTeX
    @article{bow, author = {Zhong, Zhizhen and Ghobadi, Manya and Balandat, Maximilian and Katti, Sanjeevkumar and Kazerouni, Abbas and Leach, Jonathan and McKillop, Mark and Zhang, Ying}, title = {BOW: First Real-World Demonstration of a Bayesian Optimization System for Wavelength Reconfiguration}, journal = {OFC}, year = {2021} }
    Related Publications
    ARROW: Restoration-Aware Traffic Engineering
    Z. Zhong, M. Ghobadi, A. Khaddaj, J. Leach, Y. Xia, Y. Zhang
    ACM SIGCOMM 2021
    Paper | Slides | Artifact | Video | MIT News
    Contributors