en.Wedoany.com Reported - Amazon recently announced that its developed Resilient Network Graphs (RNG) network topology architecture has begun deployment. Compared to traditional hierarchical network designs, this architecture improves performance by about one-third and energy efficiency by up to 40%.
This new architecture is based on random graph theory. Traditional data center networks use a tree-like hierarchical structure, similar to an organizational chart, where data must travel up the "chain of command" and then back down, easily creating bottlenecks at convergence points while other links may remain idle. Matt Rehder, Vice President of Global Network Engineering at AWS, stated that Amazon has been researching practical solutions for random graph network topologies over the past few years.
The concept of random graph network topology was proposed by academia as early as 2012, but the designs at that time had flaws. One approach, called "Jellyfish," relied on a completely random graph structure and separated routers from server racks, placing them centrally. However, this increased latency between servers within the same rack, introduced complex routing rules, and resulted in messy cabling, making large-scale production deployment difficult.
Inspired by other academic research, Amazon successfully achieved a random network topology by making the network not entirely random. RNG adopts a flat graph structure, mixing connections between routers through an optical device called Shufflebox, which randomizes the ports of input fibers. Additionally, the routing algorithm Spraypoint is responsible for identifying node paths. Three years ago, Amazon scholar Seshadhri Comandur and University of Washington professor Ratul Mahajan initiated related research through internal communication. With the assistance of colleagues such as Principal Applied Scientist Giacomo Bernardi, AWS became the first company to deploy a flat data center network at scale.
Internally, RNG was once called "Penrose" because the initial design involved Penrose tiles, but it was later renamed Resilient Network Graphs. This architecture has been deployed in data centers in Ireland, Germany, and Spain, with plans to cover most of the company's data centers by the end of this year. RNG is deployed on AWS's core database server network, while machine learning hardware uses the company's UltraServer network, as the latter's workloads require full bandwidth. Matt Rehder noted that the core server network can achieve more effective oversubscription. Amazon expects this technology to provide customers with better performance and reliability, while saving billions of dollars in hardware costs and reducing carbon emissions.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
