Uploaded June 2025 | Updated September 2026, 1 week ago
Modern data centers (DCs) rely on CLOS IP fabrics with EBGP to provide scalable and efficient networking. However, AI/ML workloads introduce unique challenges, demanding low-latency, lossless transport while generating high-throughput, low-entropy flows that can lead to congestion hotspots.
To address these challenges, BGP Deterministic Path Forwarding (DPF) and BGP Global Load Balancing (GLB) offer lightweight yet effective traffic engineering solutions. DPF enables deterministic IP routing, ensuring predictable flow distribution, while GLB provides sub-millisecond congestion mitigation across the fabric. These mechanisms can function independently or synergistically to enhance AI/ML traffic performance and reliability.
This talk explores the application of DPF and GLB in AI/ML fabrics, highlighting their impact on optimizing network efficiency and mitigating congestion in large-scale deployments.
Kevin Wang: Kevin Wang is a Distinguished Engineer at Juniper Networks, working in the BGP and MPLS areas for 16 years. His major innovations include BGP rib-sharding, FIB Compression, BGP Self-diagnostics, AI/ML fabric Global Load Balancing, and BGP Deterministic Path Forwarding. He actively participates in the IETF and is a coauthor of RFC-9107 as well as a few Global Load Balancing related Internet drafts. Before joining Juniper in 2008, he worked for Airvana, Axiowave Networks, Cisco, in the areas of 3G Wireless, IP/MPLS protocols, and Content Delivery.
Michal Styszynski: Michal is part of the Product Management team at Juniper Networks' cloud-ready data center business unit. He joined Juniper Networks over 10 years ago. Before his current PLM role, he had also worked in technical marketing and product consulting, focusing on Data Center and Storage networking projects for major telcos and large enterprises. Before Juniper, Michal also worked for about 10 years at Orange, former France Telecom R&D.
Speaker Kevin Wang - Juniper Networks
Michal Styszynski - Juniper Networks
Modern data centers (DCs) rely on CLOS IP fabrics with EBGP to provide scalable and efficient networking. However, AI/ML workloads introduce unique challenges, demanding low-latency, lossless transport while generating high-throughput, low-entropy flows that can lead to congestion hotspots.
To address these challenges, BGP Deterministic Path Forwarding (DPF) and BGP Global Load Balancing (GLB) offer lightweight yet effective traffic engineering solutions. DPF enables deterministic IP routing, ensuring predictable flow distribution, while GLB provides sub-millisecond congestion mitigation across the fabric. These mechanisms can function independently or synergistically to enhance AI/ML traffic performance and reliability.
This talk explores the application of DPF and GLB in AI/ML fabrics, highlighting their impact on optimizing network efficiency and mitigating congestion in large-scale deployments.
Kevin Wang: Kevin Wang is a Distinguished Engineer at Juniper Networks, working in the BGP and MPLS areas for 16 years. His major innovations include BGP rib-sharding, FIB Compression, BGP Self-diagnostics, AI/ML fabric Global Load Balancing, and BGP Deterministic Path Forwarding. He actively participates in the IETF and is a coauthor of RFC-9107 as well as a few Global Load Balancing related Internet drafts. Before joining Juniper in 2008, he worked for Airvana, Axiowave Networks, Cisco, in the areas of 3G Wireless, IP/MPLS protocols, and Content Delivery.
Michal Styszynski: Michal is part of the Product Management team at Juniper Networks' cloud-ready data center business unit. He joined Juniper Networks over 10 years ago. Before his current PLM role, he had also worked in technical marketing and product consulting, focusing on Data Center and Storage networking projects for major telcos and large enterprises. Before Juniper, Michal also worked for about 10 years at Orange, former France Telecom R&D.
Speaker Kevin Wang - Juniper Networks
Michal Styszynski - Juniper Networks
