TL;DR
Amazon has successfully built and deployed a flat data center network using a novel RNG design based on random graphs, achieving significant improvements in efficiency and resilience. The project moved from research to real-world implementation in 2024 and is now the default for new data centers.
Amazon has completed the deployment of its first large-scale flat data center network based on a novel random graph architecture called RNG, marking a significant shift from traditional hierarchical topologies. This development, confirmed by AWS sources, aims to improve resilience, reduce costs, and enhance throughput across its global infrastructure.
Researchers at Amazon, led by principal scientist Giacomo Bernardi and scholar Ratul Mahajan, developed the RNG network design after years of theoretical and simulation work. The design replaces traditional hierarchical fat-tree topologies with a flat, randomized graph structure that exploits properties of expanders to optimize routing and cabling.
In 2024, Amazon built the first RNG-based data center near Dublin, Ireland, demonstrating the practical viability of the approach. By 2025, two additional RNG data centers were completed in Germany and Spain. These deployments showed that RNG networks used 69% fewer routers, achieved 33% higher throughput, reduced power consumption by 40%, and lowered operating costs by 27%, compared to conventional fat-trees.
Amazon’s team developed specialized routing schemes like Spraypoint, passive optical devices called ShuffleBox for cabling, and software models to predict network performance, enabling deployment at scale without disrupting existing infrastructure. The RNG design is now the default for most new Amazon data centers worldwide, as announced internally in early 2026.
Impact of RNG on Data Center Infrastructure
The deployment of RNG networks at Amazon signifies a major evolution in data center design, emphasizing simplicity, resilience, and cost-efficiency. By replacing complex hierarchical topologies with flat, randomized graphs, Amazon reduces hardware needs, enhances fault tolerance, and simplifies cabling, which can lead to broader industry shifts towards similar architectures.
This move could influence data center standards globally, encouraging other cloud providers to adopt similar flat, scalable network designs that leverage the mathematical properties of expanders and random graphs.
FLYPROFiber- 30M OS2 Fiber Patch Cable LC to LC, Length Options: 0.2m-200m, 1G/10GB Single Mode Duplex, 9/125um SMF Fiber Optic Cable Cord LSZH 30Meter(98ft)
📡High Rated OS2 9/125μm Fiber and Cladding, which is Insensitive to bending,Easy peeling,Easy welding,ensures small optical loss and…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Historical Evolution of Data Center Networking Architectures
Traditional data center networks have relied on hierarchical topologies like fat-trees and Clos networks, which, while scalable, are complex and cable-intensive. Research into optimal network routing dates back to the 1970s, with mathematicians exploring expanders—graphs with strong connectivity properties. In the 2000s, industry efforts incorporated randomness into structured topologies, such as the VL2 architecture, to improve load balancing and performance.
In 2012, the Jellyfish model introduced the idea of using random graphs for data centers, but practical challenges in routing and cabling limited its adoption. Recent mathematical advances demonstrated that purely random graphs could match the properties of optimal expanders, leading to renewed interest. Amazon’s recent research and deployment of RNG networks build directly on these theoretical insights, translating them into operational infrastructure.
“Replacing structure with randomness has shown dramatic improvements in our data centers, proving the concept works in real deployments.”
— Ratul Mahajan, Amazon Scholar
“If you want to demonstrate that it works, go build it in an actual data center.”
— Matt Rehder, VP of Network Engineering
S8550-32C, 32 x 100Gb QSFP28, L3 Managed Switch, Front-to-Back Airflow, MLAG | VXLAN | EVPN | BGP | OSPF
1+1 Hot-swappable Power Supplies, 3+1 Smart Fans, 6.4 Tbps High Throughput; Support OSPF/OSPFv3, BGP4/BGP4+, ISIS, BFD, LACP, VRRP,…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Unresolved Questions About RNG Network Scalability
While initial deployments are promising, it remains unclear how RNG networks will perform at even larger scales or under diverse workload conditions. Long-term operational stability, maintenance, and compatibility with existing infrastructure are still being evaluated. Industry adoption beyond Amazon has not yet been confirmed, and the full impact on global data center standards is uncertain.
XGSPON ONU Stick Optical Transceivers,with MAC SFP+ TX-9.95G/RX-9.95G TX-1270nm/RX-1577nm SC/APC 20km DDM,This Product is Applicable to The Mellanox Series 10G Switch
This product is applicable to the Mellanox Series 10G switch models: Mellanox ConnectX-3;Mellanox Connectx-4 Switch. If you plan…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for RNG Data Center Networks
Amazon plans to continue deploying RNG-based networks in more data centers worldwide, monitoring performance and resilience over time. Further research aims to refine the models, improve automation tools for deployment, and explore integration with other network architectures. Industry observers will watch for potential adoption by other cloud providers or data center operators, as well as for updates on long-term operational results.
Tecmojo 42U Server Rack Network Cabinet 39.4" D x 23.6" W, Locking Data Cabinet Enclosure for 19" Server, Networking, AV & IT Equipment, Glass Door, Black
Superior Load Capacity: 42U server rack supports up to 1800lbs,max mountable depth is 34.25in, ideal for heavy IT…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is the RNG network architecture?
It is a flat, randomized graph-based network design that replaces traditional hierarchical topologies, leveraging properties of expanders for improved resilience and efficiency.
How does RNG improve data center performance?
RNG reduces the number of routers needed, increases throughput, lowers power consumption, and simplifies cabling, resulting in cost savings and better fault tolerance.
Is RNG deployment limited to Amazon?
Currently, it is primarily implemented at Amazon, but the underlying research and promising results suggest wider industry interest may develop.
What challenges remain with RNG networks?
Long-term operational stability, scalability, and integration with existing infrastructure are still being evaluated, along with potential industry adoption hurdles.
When will RNG networks be available for other companies?
There is no confirmed timeline; Amazon’s deployments are ongoing, and broader adoption depends on further research, validation, and industry interest.
Source: Hacker News
