Physical Network Design at Scale
summary
Loom automates end-to-end physical network design: Planogram places thousands of devices onto racks and failure domains using lexicographic optimization (four ranked objectives: connectivity, failure spreading, row similarity, rack efficiency), then a fiber design engine materializes millions of fiber strands into trunk, breakout, and shuffle cable patterns with mandated path diversity. The stack combines ILP + RL + graph reduction to produce build-ready designs across four regions and 20+ buildings, cutting design time from months to hours.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:53Automating network design pipeline
- 3:38Placement problem challenges
- 6:48Balancing placement objectives
- 9:16Planagram system engines
- 11:03Fiber infrastructure after placement
- 13:53Fiber design rules
- 16:24Real-world adoption gains
- 17:29Proving designs better
- 18:13Lessons from automation
- 19:14Future directions
related talks
21mLarry: Practical Network Reconfigurability in the Data Center
28mLoom: Flexible and Efficient NIC Packet Scheduling
23mMinimal Rewiring: Efficient Live Expansion for Clos Data Center Networks
17mMeta’s Network Journey to Enable AI
16mLEOCraft: Towards Designing Performant LEO Networks
24mNetBlocks: Staging Layouts for High-Performance Custom Host Network Stacks
30mHow We Built an Automated Data Center Network in Less Than 6 Months
41mHow to Build a Network Automation Stack From The Ground Up