TL;DR
Amazon has successfully implemented RNG-based flat data center networks at scale, achieving significant efficiency and resilience improvements over traditional fat-tree designs. The deployment began in 2024 and is now the standard for new data centers globally.
Amazon has deployed Resilient Network Graphs (RNG) in its data centers, replacing traditional hierarchical fat-tree networks with a flat, randomized architecture that offers improved efficiency and resilience. This marks a significant shift in data center networking at scale, with the deployment now the standard for new Amazon facilities globally.
Researchers at Amazon, led by principal scientist Giacomo Bernardi and scholar Ratul Mahajan, developed RNG-based network designs that leverage randomness and advanced routing schemes to create flat, scalable data center networks. After extensive simulation and experimental validation in a data center near Dublin in 2024, Amazon expanded the deployment to data centers in Germany and Spain in 2025. The RNG architecture reduces the number of routers by 69%, increases throughput by 33%, cuts power consumption by 40%, and lowers operating costs by 27%, compared to traditional fat-tree networks.
Unlike hierarchical networks, RNG provides proportional and predictable degradation in case of failures, with no single point of critical failure. The design employs specialized routing schemes like Spraypoint and passive optical devices called ShuffleBoxes to facilitate cabling and traffic distribution, all compatible with existing Amazon hardware. The projects were guided by models that predict performance, enabling deployment validation before physical installation.
Impact of RNG on Data Center Design and Operations
The adoption of RNG-based flat networks by Amazon represents a major advancement in data center architecture, offering substantial improvements in efficiency, resilience, and operational costs. This development could influence industry standards, encouraging other cloud providers to explore similar architectures for scalable, reliable infrastructure.
TP-Link TL-SG105, 5 Port Gigabit Unmanaged Ethernet Switch, Network Hub, Ethernet Splitter, Plug & Play, Fanless Metal Design, Shielded Ports, Traffic Optimization
- Number of Ports: 5 Gigabit Ethernet Ports
- Network Speed: Supports 10/100/1000Mbps
- Auto Features: Auto Negotiation and MDI/MDIX
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Evolution from Hierarchical to Randomized Networks
Traditional data center networks relied on hierarchical fat-tree topologies inspired by Clos switches, which became increasingly complex and cable-intensive as cloud demands grew. Research into optimal routing and network expanders dates back to the 1970s, with recent mathematical insights showing that random graphs can match the performance of complex explicit designs. Early work like Jellyfish in 2012 introduced the concept of random graph topologies, but practical scaling and routing remained challenging until now. Amazon’s recent research and deployment build on these theoretical foundations, translating them into operational networks at scale.
“Randomized network designs, once purely theoretical, are now practical solutions that outperform traditional hierarchical architectures at scale.”
— Ratul Mahajan, Amazon Scholar
“The success of our RNG data centers demonstrates that innovative network topology can deliver tangible benefits without overhauling existing hardware.”
— Matt Rehder, VP of Network Engineering
YIENFBEV LC Fiber Optic Adapter – LC to LC Duplex Singlemode Coupler – 10 Pack – Blue
- Adapter Type: LC to LC Duplex Fiber Optic Adapter
- Connector Material: Ceramic Sleeve for Connection
- Locking Mechanism: Connectors Lock Securely
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Remaining Questions on RNG Deployment and Scalability
It is not yet clear how RNG networks will perform under extreme failure conditions or in different operational environments beyond Amazon’s initial deployments. Long-term reliability, maintenance, and upgrade procedures for RNG-based networks are still being evaluated, and the broader industry adoption remains uncertain pending further validation.
Mikrotik CCR1016-12G Routerboard-Cloud Core Router
- Processor: 16-core CPU for high performance
- Ethernet Ports: 12 Gigabit Ethernet ports
- Display: Color touchscreen LCD
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for RNG in Data Center Networking
Amazon plans to continue scaling RNG deployments across its global data centers, refining the models and tools used for design and diagnostics. Industry observers expect other cloud providers and data center operators to evaluate RNG architectures, potentially leading to broader adoption if performance and reliability are confirmed in diverse scenarios. Amazon may also publish detailed performance data and best practices to facilitate wider industry understanding.
2 Pack 3dB SC Fiber Optic Attenuator, Single Mode Fixed in-Line Fiber Attenuator, SM SC/UPC Male to Female, dB Options: 3dB, 5dB, 7dB, 10dB
- Type: 3dB SC/UPC Fiber Optic Attenuator
- Pack Size: 2 Pack
- Connection: Male to Female SC/UPC
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is RNG in data center networks?
RNG, or Resilient Network Graphs, are flat, randomized network architectures that use mathematical principles to optimize routing, cabling, and resilience, replacing traditional hierarchical topologies.
How does RNG improve over traditional fat-tree networks?
RNG reduces hardware requirements by 69%, increases throughput by 33%, lowers power consumption by 40%, and offers more predictable failure behavior, making networks more efficient and resilient.
Are RNG networks ready for widespread industry adoption?
While Amazon has successfully deployed RNG at scale, broader industry adoption depends on further validation of long-term reliability and operational practices. It is still an emerging approach.
What challenges remain for RNG deployment?
Challenges include ensuring robustness under extreme failures, developing scalable management tools, and integrating RNG with existing infrastructure and operational workflows.
Source: Hacker News
