Amazon has introduced a new routing architecture for its AWS data centers, known as Resilient Network Graphs (RNG), which significantly enhances throughput and reduces energy costs. This innovative design aims to replace the aging traditional ‘fat tree’ topology that has been standard in data centers for years.
The RNG architecture, implemented since April 2026, is said to improve throughput by 33% while decreasing the required number of switch-routers by 69%. As a result, Amazon claims that it could potentially cut down network infrastructure electricity usage by 40%. This transition promises to enhance service reliability without requiring customers to alter their existing code.
Traditional fat tree routing, established in the 1990s for supercomputing, has continued to be the go-to solution due to its scalability. However, it leads to increased complexity and congestion issues as data centers expand. The hierarchical structure demands extensive switch and cabling, pushing designers to compromise on costs, adversely affecting performance.
The RNG, influenced by random graph theory, proposes a non-hierarchical mesh topology to facilitate better efficiency. Research initiatives like the Jellyfish project in 2012 have long suggested the potential advantages of a random graph approach, but complexities in cabling and node routing tables have hindered practical applications.
Amazon’s researchers have developed a hybrid routing algorithm called Spraypoint, which integrates elements of both RNG and traditional fat tree designs. This quasi-random network sprays traffic to neighboring switches while employing conventional shortest path algorithms as packets approach their destination. A novel device, named ShuffleBox, consolidates the intricate wiring required for random connections, simplifying overall data center design.
While the performance improvements claimed by Amazon have yet to be independently verified, the architecture is already live in several data centers, with many more planned for deployment. The first RNG network has been operational near Dublin, marking a pivotal shift in AWS infrastructure.
Industry experts commend the RNG for its potential in addressing the increasing energy demands and sustainability challenges faced by data centers. However, its proprietary nature may limit widespread adoption, as most hyperscale clients might not bear the financial burden of implementing such advanced systems. Consequently, Amazon has indicated that RNG will primarily be used in new data centers, making it unlikely for competitors to adopt this cutting-edge technology immediately.
For more on the Resilient Network Graphs and their implications in the industry, visit Amazon Web Services.
