Case study: AI-Engine Driven Cross-Domain Orchestration implementation in Dynamic Optical Networks

The cross-domain orchestration to efficiently manage end-to-end optical links across multiple network domains over a 500km field trial testbed can be demonstrated, which consists three network domains with real-time transponders. This case study includes several innovations: i) a machine learning-based flex-grid multi- channel QoT estimator (FM-QOT) to enable link- performance aware routing; ii) cross-domain joint network rerouting in response to detected performance degradation, which is achieved through LSTM-based performance analysis or the predicted cross-channel impact based on the FM-QOT for the defined scenarios; iii) deployed AI- engine to manage multiple machine learning applications. Enabled by the developed FM-QOT, performance-aware routing achieves an average link performance improvement of approximately 20% across 10k services in the transport domain. When network rerouting is necessary with the detected performance degradation, the cross-domain orchestrator redirects edge traffic through different transport links to improve network resilience. This rerouting process utilizes the FM-QOT hosted by an AI engine with time series databases and multiple domain controllers to complete end-to-end traffic rerouting within 800 ms. Two scenarios are demonstrated for network rerouting: one involving link performance degradation when adding new channels through cross-channel impact analysis provided by the FM-QoT, and another using LSTM-based link performance analysis. The cross-domain management show-cases a potential approach for future converged networks with joint optimization.

Figure 1 shows the developed field-trial multi- domain testbed with the cross-domain orchestrator. The field trial testbed includes two edge network domains consisting of network servers with 100G network interface cards and one transport domain with up to 12 real-time transponders for cross-domain end-to-end optical connections. The two edge network domains are implemented based on the 5G UK testbed with deployed fibre links connecting multiple sites including We The Curious (WTC), Mshed, Watershed, and NSQI in Bristol, UK. The edge network traffic rerouting is achieved via controlling fibre switches using the REST API to connect 100G clients to different transport transponders. The transport network domain consists of a 493km loop link with deployed fibre over part of the UK National Dark Fibre Facility (NDFF) stretching from the University of Bristol (UoB) to the Powergate via Bradley Stoke, Froxfield, and Reading, as shown in Fig. 1(b).

Initially, end-users employ 100G QSFP modules, designed for 10km long-range transmissions, to send data via the established fibre links. This data traffic is then aggregated at the aggregation nodes (ANs) using optical data center interconnection transponders: Voyagers and Teraflexs, as shown in Fig. 1(a). Each Voyager unit contains 4 coherent network transponders and 8 client ports and the Teraflex unit comes with 2 coherent network transponders and 12 client ports. All client ports across both transponders support 100G QSFP modules. The Voyager transponder can handle traffic of up to 200G using real-time 32Gbaud Nyquist PM-16QAM optical signals. On the other hand, the Teraflex coherent transponder has varying capacities for different traffic loads. They can support 200G, 300G, and 400G by utilizing real-time Nyquist PM-16QAM optical signals with baud rates of 32G, 47G, and 63G respectively.

On top of the physical connections, each domain has its own domain controller. Especially, an AI engine that hosts ANN-based FM-QoT and LSTM algorithms is deployed in the transport network domain. The cross-domain orchestrator is developed to handle cross-domain configurations and rerouting.

Fig. 1: Field-trial testbed. (a) AI-engine-based cross-domain orchestrator in the dynamic optical network. (b) Field-trail physical layer configuration.

Funding information: This work has been funded by European Union’s Horizon RIA project ALLEGRO (No.101092766)

For further information: contact Dr. Rui Wang, ndff@ee.ucl.ac.uk

Published: 15 Feb 2024