Layer 3 networking with BGP in hyperscale DCs
summary
Proposes replacing large L2 domains with an all-L3 fabric using BGP, with FRRouting (FRR) as the control plane and run an OVN daemon on each hypervisor to advertise /32 host routes (or /128 IPv6) directly to ToR switches over ECMP and /31 point-to-point links for BGP sessions. Notes Neutron’s dynamic-routing only advertises tenant subnets, not provider host routes; calls out limitations: no overlapping IPs, SRv6/SRv6 offloads and SR-IOV/DPDK paths break, and performance/scale testing is still needed.
AI-written summary · about
chapters (12)
- 0:00Intro
- 0:16Agenda
- 0:43Data center evolution
- 2:48Layer 2 networking
- 6:45What is BGP
- 8:13Introduction
- 9:37Openstack use cases
- 11:42Openstack walkthrough
- 15:01Limitations
- 17:42Use cases
- 19:26Design
- 20:34Next steps
related talks
32mLeveraging open source dynamic routing to implement a spine/leaf in datacenter
30mHow we integrated L3 services into one of the biggest OpenStack Public Clouds - OVHcloud case study
19mCrossing the border - VM goes on a BGP adventure
29mLiving on the Edge with DCN Networking
34mHyperRouter: Lessons Learnt from Building an L4 Load...
39mUsing BGP at OpenStack to interconnect workloads across clouds
27mLeveraging Cilium and SRv6 for Telco Networking
23mAIN: A Blueprint for an All-IP Data Center Network