Running BGP in Data Centers at Scale
summary
Facebook describes a BGP-based data-center routing design that scales across hundreds of pods by reusing ASN blocks, hierarchical route summarization to cut hardware FIB and control-plane load, and a policy system that installs predefined backup paths so failures switch locally without waiting for global reconvergence. An in-house BGP agent plus automated test-and-deploy pipeline pushes bi-weekly agent updates and policy changes. Design runs fleet-wide for 5+ years and incidents over a two-year window drive iterative policy and implementation fixes.
AI-written summary · about
chapters (5)
related talks
38mFacebook's Data Center Fabric
41mImpact of Network Automation
19mEngineering Facebook's Data Center Network: Challenges and Limitations
29mUsing the Internet as Your Load-Balancer
46mLessons learned from monitoring a large production network
12mxBGP: Faster Innovation in Routing Protocols
39mUsing BGP at OpenStack to interconnect workloads across clouds
22mLayer 3 networking with BGP in hyperscale DCs
