ICE seminar: « Control within Future AI-Datacenter Fabrics »
Abstract
Datacenter control implementing forwarding & service topology, and routing i.e., how IP traffics are steered, are evolving because large, bursty, east–west traffic for AI-Application workloads must be carried by dynamic and high-throughput flows at different scales: scale-up, scale-out and scale-across.
IP control deployed for stable WAN-style connectivity—cannot meet latency, agility, cost, or security requirements of flows in and between AI-DC fabrics. These are the reasons why DC and WAN are shifting toward programmable fabrics, overlays/underlays, centralized hierarchical control, latency measurement-controlled paths.
This seminar presents the new software defined network control architectures and its engines addressing the new routing requirements following two complementary approaches: (i) inside Datacenters driven by the limits of Layer 2/RSTP that is not scalable for east-west intra-DC traffics and (ii) between datacenters driven by the limits of BGP with Traffic-Engineering extensions and the restrictions of stretched Layer 2 protocols.
With respect to these challenges, the seminar introduces how (i) Routing In Fat Tree (RIFT) protocols can be used for intra-DC networks characterized by regular topologies with a high degree of connectivity, defined directionality, and large scale and (ii) BGP Link State-Shortest Path First ((BGP-LS-SPF) routing protocol are modeled and can be designed for interconnected AI-Datacenter Fabrics.
Bio
Dominique Verchere has been Principal Research Architect at Network Research Architecture (NAR) Lab of Bell Labs, part of the Strategy & Technology of NOKIA & he is Expert on IP & Optical Network Control and Service Management. He holds an Electrical Engineering Bachelor of Sciences degree in Physics and Measurements, Master of Sciences degree on Material Physics (Grenoble University), Master’s degree in computer science and PhD degree in Computer Science (Pierre & Marie University, now Sorbonne University) in Telecom Systems with stochastic modelling, queuing theory and performance evaluation. He had continuously researched and applied breakthrough concepts on Software Defined Networking (SDN) Control for Flexible Wavelength Division Multiplexing (DWDM) Networks and Digital Optical Transport Networks (OTN). He proved automation concepts on open disaggregated transport networks for datacenter interconnects to deliver secured and sustainable Cloud Services.
He produced several contributions to Open Networking Foundation especially to ODTN project (Open Disaggregated Transport Networks) evaluating ONOS SDN-C, CORD (Central Office Re-architected as a Datacenter) platform, OpenDaylight SDN-C, Transport-API realizing proof-of-concepts with Telecom Network Service Providers (Telefónica, NTT, Orange) on transport network service enablement and multi-layer resource control. Recently with respect to quantum-secured networks, he has investigated different key allocation strategies to enhance the performance of key management systems in Quantum Key Distribution Networks (QKD networks) and to evaluate quantum key allocation success rate and key delivery delay. Within NAR Lab, he is working on Self-Evolving Network Concepts within Agentic AI Framework for Autonomous Transport Networks and on routing protocols for future datacenter fabrics built on IP & Optical Networks.