TRANSNET researchers produce comprehensive topology dataset to accelerate benchmarking studies in optical networks.

TRANSNET researchers produce comprehensive topology dataset to accelerate benchmarking studies in optical networks.

Introduced in a new paper published with Journal of Optical Communications and Networking, Topology Bench is a powerful research resource to address complex optimisation challenges in optical networking, including physical topology design and resource allocation problems.

Algorithm testing has traditionally relied on subjective topology selection, leading to biases and limited network representation. Prior work lacked systematic graph-theoretical analysis, leaving gaps in understanding the structural, spatial, and spectral properties that affect network performance. Furthermore, the small scale and lack of diversity in available datasets constrained the range of configurations possible to test algorithms across. Without a unified repository combining real-world and synthetic topologies with essential data, researchers lacked a comprehensive benchmarking resource.

Topology Bench significantly advances benchmarking for wavelength-routed optical networks through three critical contributions. First, it delivers the most extensive open-access dataset to date, comprising 105 georeferenced real-world core network topologies and 270,900 synthetically generated topologies. Second, it integrates structural, spatial, and spectral graph-theoretical metrics to rigorously characterise the properties of optical networks. Finally, it introduces a robust framework for systematically selecting representative topologies by employing unsupervised machine learning techniques, clustering real-world networks into objectively defined groups.

These contributions allow researchers to choose a diverse set of networks – based on their structural, spatial and spectral properties - to benchmark network algorithms. The large set of synthetic networks provide a wide variety of networks to use for machine learning training in the future.

About the paper

The work, published in Journal of Optical Communications and Networking, was completed by TRANSNET researchers at UCL and the University of Cambridge.

"Topology Bench: systematic graph-based benchmarking for core optical networks," J. Opt. Commun. Netw.  17, 7-27 (2025).

Authors: Robin Matzner, Akanksha Ahuja, Rasoul Sadeghi, Michael Doherty, Alejandra Beghelli, Seb Savory, and Polina Bayvel.

Read the full paper here.