Fast WDM provisioning with minimal probing: the first field experiments for DC exchanges

Hideki Nishizawa, Toru Mano, Thomas Ferreira de Lima, Yue-Kai Huang, Zehao Wang, Wataru Ishida, Masahisa Kawashima, Ezra Ip, Andrea D'Amico, Seiji Okamoto, Takeru Inoue, Kazuya Anazawa, Vittorio Curri, Gil Zussman, Daniel C. Kilper, Tingjun Chen, Ting Wang, Koji Asahi, Koichi Takasugi

TL;DR

Researchers have demonstrated a new method for lightning-fast provisioning of optical network paths in metro-scale Data Center Exchanges (DCX). By using a "minimal probing" technique—measuring signal quality link-by-link rather than end-to-end—and an open, containerized software architecture, they successfully established high-capacity WDM links across Manhattan field fibers in under 6 minutes with high accuracy.

What Problem Does This Paper Address?

As cloud services and mobile edge computing (MEC) expand, we need to connect distributed data centers (DCI) and exchanges (DCX) quickly and flexibly. Traditionally, setting up a Wavelength Division Multiplexing (WDM) link requires measuring the Quality of Transmission (QoT) for a specific path and wavelength.

In a complex metro network with many-to-many connections and diverse hardware (different vendors, modulators, and frequency ranges), this "measure everything" approach is too slow. Furthermore, network operators often lack the specific internal characteristics of a user's transceiver (TRx), making it difficult to predict how a signal will degrade over a specific route. The industry lacks a unified way to manage these multi-vendor environments while maintaining fast, automated restoration and provisioning.

Methods

The authors propose a two-pronged solution:

  1. Link-by-Link GSNR Estimation: Instead of measuring the end-to-end (EtE) Generalized Signal-to-Noise Ratio (GSNR) for every possible route, they measure the GSNR of individual links using a "probe" channel at a convenient wavelength. They then use a mathematical model based on Gaussian noise approximation to sum the reciprocal GSNRs of these links (plus the transceiver's own internal noise) to accurately predict the total end-to-end signal quality.
  2. Open Software Architecture: They developed a Linux-based transmission software stack using the Goldstone Network Operating System (NOS) and the Transponder Abstraction Interface (TAI). Crucially, they use Docker containers to allow a carrier's controller to interact with a user's transceiver. This "Carrier Agent" container can extract vital hardware characteristics (stored as JSON files) without requiring the user to disclose proprietary source code, enabling the controller to calculate the optimal transmission mode (e.g., 400G 16QAM vs. 200G QPSK) automatically.

Datasets

While not a traditional machine learning paper, the study relies on real-world experimental data from:

  • NSF COSMOS Testbed: A city-scale programmable optical/wireless testbed in Harlem, New York City.
  • Manhattan Field Fibers: Real-world dark fiber loops (32 km and 8 km) provided by Boldyn and CrownCastle, connecting Columbia University to other NYC data hubs.
  • Hardware: Commercial 400G CFP2-DCO transceivers from Fujitsu and Lumentum, utilizing both Indium Phosphide (InP) and Silicon Photonics (SiP) modulators.

Key Insights and Ideas

  • The "Sum of Reciprocals" Works: The paper confirms that EtE GSNR can be predicted by summing the reciprocal noise contributions of the transceivers, amplifiers (ASE noise), and fiber nonlinearities.
  • Wavelength Independence in Metro Scales: In metro-scale networks (up to ~200km), the wavelength-dependent error is surprisingly small (~0.2 dB). This means a probe at one end of the C-band is a reliable proxy for the rest of the band, significantly simplifying the probing process.
  • Bridging the "Back-to-Back" Gap: Standard network models (like YANG) often miss the "back-to-back" characteristics of transceivers. By using a containerized agent to pass this data via JSON, the authors solved a major interoperability hurdle for multi-vendor DCX.
  • Speed is the Killer App: By automating the "margin design" and configuration, they moved from manual, slow provisioning to a fully automated process that takes minutes, not hours or days.

Key Results

  • High Accuracy: The difference between the estimated GSNR (via link-by-link probing) and the actual measured end-to-end GSNR was only 0.6 dB.
  • Minimal Wavelength Error: The error caused by using a probe at a different wavelength than the operational signal was only about 0.2 dB.
  • Rapid Provisioning: The entire process—from gathering transceiver info to measuring links, selecting the optimal mode, and establishing the path—was completed within 6 minutes (351 seconds).
  • Successful Field Trial: They successfully provisioned a 400G DP-16QAM path for a short route and two 200G DP-QPSK paths for a longer, more degraded route automatically.

Limitations and Future Work

  • C-Band Focus: The experiments were conducted in the C-band. The authors note that in C+L multi-band systems, wavelength-dependent effects like Raman crosstalk and amplifier gain profiles will become more complex, requiring more advanced estimation techniques (like transfer learning).
  • Form Factor Diversity: The current implementation focused on CFP2-DCO modules. Future work needs to accommodate other popular form factors like QSFP-DD, which may require utilizing different management interfaces like SONiC.
  • Nonlinearities: While fiber nonlinear effects were small in this metro setup, they would play a much larger role in long-haul systems, potentially requiring more sophisticated probing or modeling.

Summary updated

Abstract

We propose an approach to estimate the end-to-end GSNR accurately in a short time when a data center interconnect (DCI) network operator receives a service request from users, not by measuring the GSNR at the operational route and wavelength for the End-End optical path but by simply applying a QoT probe channel link by link, at a convenient wavelength/modulation-format for measurement. Assuming connections between coherent transceivers of various frequency ranges, modulators, and modulation formats, we propose a new device software architecture in which the DCI network operator optimizes the transmission mode between user transceivers with high accuracy using only standard parameters such as Bit Error Rate. In this paper, we first experimentally built three different routes of 32 km/72 km/122 km in the C-band to confirm the accuracy of this approach. For the operational end-to-end GSNR measurements, the accuracy estimated from the sum of the measurements for each link was 0.6 dB, and the wavelength-dependent error was about 0.2 dB. Then, using field fibers deployed in the NSF COSMOS testbed (deployed in an urban area), a Linux-based transmission device software architecture, and coherent transceivers with different optical frequency ranges, modulators, and modulation formats, the fast WDM provisioning of an optical path was completed within 6 minutes (with a Q-factor error of about 0.7 dB).
BibTeX
@article{nishizawa2024fast,
  author = {Hideki Nishizawa and Toru Mano and Thomas Ferreira de Lima and Yue-Kai Huang and Zehao Wang and Wataru Ishida and Masahisa Kawashima and Ezra Ip and Andrea D'Amico and Seiji Okamoto and Takeru Inoue and Kazuya Anazawa and Vittorio Curri and Gil Zussman and Daniel C. Kilper and Tingjun Chen and Ting Wang and Koji Asahi and Koichi Takasugi},
  title = {{Fast WDM provisioning with minimal probing: the first field experiments for DC exchanges}},
  journal = {J. Opt. Commun. Netw.},
  year = {2024},
  doi = {10.1364/jocn.505729},
}

Loading related content...

Figure