AI Briefing
KO

AWS Data Center Networks Adopt 'Flat' Structure Instead of 'Fat Tree'

·2026.05.28 19:30

Key point

AWS has introduced a quasi-random flat network to replace the existing fat tree structure, improving efficiency and resilience.

Details

Existing data center routing mainly uses the hierarchical fat tree structure. However, this approach incurs overhead from additional router layers, and suffers from reduced efficiency and stability due to bottlenecks and single points of failure at the top-level routers.

In theory, a flat network where routers are directly connected is the most ideal. However, random connections require enormous computational resources for path calculation, and complex cabling issues made actual implementation impossible.

To solve this, AWS developed RNG (Resilient Network Graphs), a quasi-random network topology. This was combined with ShuffleBox, a passive optical component, to solve the physical cabling problem of flat networks.

The routing algorithm Spraypoint works as follows:

  • The source router performs spraying, randomly distributing traffic to neighboring nodes
  • Traffic is guided through waypoints of each destination router
  • The rings structure around the destination is used to prevent traffic congestion

Through this design, AWS data centers achieved the following results:

  • 69% reduction in number of routers
  • Up to 33% improvement in throughput
  • Expected 40% reduction in network equipment power consumption

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.