Physical Network Design at Scale
summary
Meta describes Loom, an end-to-end system that automates physical network design: Planogram places every network device to a precise rack and row using an ILP + reinforcement learning decomposition and lexicographic objectives, then a fiber module materializes connectivity into trunk, breakout and shuffle cables. Results: deployed across 4 regions, 20+ buildings and 100 concurrent design variants, initial device+fiber allocation drops from months to days, validation drops to minutes, and the system outputs installation instructions and millions of strands modeled and BOMs.
AI-written summary · about
chapters (5)
- 0:00Introduction
- 2:32Placement automation module
- 10:57Fiber design automation
- 16:34Real-world impact and future
- 19:47Q&A
related talks
24mWiring the Planet: Scaling Meta's Global Optical Network
21mChange Management in Physical Network Lifecycle Automation
17mMeta’s Network Journey to Enable AI
21mExperiences with Modeling Network Topologies at Multiple Levels of Abstraction
18mNetwork Workflow Automation at Scale
23mMinimal Rewiring: Efficient Live Expansion for Clos Data Center Networks
30mUsing Python to design, configure, and measure large-scale networks
19mDeveloping, Deploying, Operating In-House Network Switches at a Massive Scale
