--- title: Goodbye, Leaf-and-Spine Networks? « ipSpace.net blog description: An AWS engineering paper claims a quasi-random graph (RNG) network fabric achieves 33% better throughput with 69% fewer routers, 27% cost savings, and 40%... canonical: https://daily.dev/posts/goodbye-leaf-and-spine-networks-ipspace-net-blog-qkpkkzvvc twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Goodbye, Leaf-and-Spine Networks? « ipSpace.net blog | daily.dev og:description: An AWS engineering paper claims a quasi-random graph (RNG) network fabric achieves 33% better throughput with 69% fewer routers, 27% cost savings, and 40%... og:url: https://daily.dev/posts/goodbye-leaf-and-spine-networks-ipspace-net-blog-qkpkkzvvc og:image: https://api.daily.dev/og/posts/QKpkKZVVC.png og:image:alt: Goodbye, Leaf-and-Spine Networks? « ipSpace.net blog og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Goodbye, Leaf-and-Spine Networks? « ipSpace.net blog **[ipSpace.net Blog](https://daily.dev/sources/ipspace)** · 8 min read · 0 upvotes · 0 comments ## Summary An AWS engineering paper claims a quasi-random graph (RNG) network fabric achieves 33% better throughput with 69% fewer routers, 27% cost savings, and 40% lower power usage compared to traditional leaf-and-spine designs. The author critically examines these claims, arguing the approach is a rediscovery of what Plexxi attempted years ago using ShuffleBoxes (prewired partial mesh) instead of spine switches, combined with unequal-cost multipathing. The author contends that in a true apples-to-apples comparison with a non-oversubscribed leaf-and-spine fabric, the AWS design cannot achieve higher throughput — any apparent gains likely stem from comparing against an oversubscribed or suboptimally load-balanced baseline. The power savings are plausible since ShuffleBoxes are passive elements. The approach may be valuable for hyperscale fabrics with tens of thousands of switches, but is unlikely to be relevant for enterprise data centers. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What is the AWS quasi-random graph (RNG) network fabric and how does it differ from leaf-and-spine? The AWS RNG fabric replaces spine switches with a prewired partial mesh device called a ShuffleBox, connecting leaf (ToR) switches directly via split uplink lanes. Traffic uses unequal-cost multipathing across multiple paths, including relay hops through other leaf switches. The design is mathematically a quasi-random expander graph, not truly random. It requires more ToR uplinks than a comparable leaf-and-spine fabric to achieve the same oversubscription ratio. _Engineers evaluating hyperscale fabric alternatives track papers like this on daily.dev before committing to a design direction._ ### Can a partial mesh or expander graph network fabric achieve higher throughput than a non-oversubscribed leaf-and-spine fabric? No — in a true apples-to-apples comparison, a partial mesh cannot exceed the throughput of a non-oversubscribed leaf-and-spine fabric. Traffic in a partial mesh may traverse more than two leaf uplinks (using relay hops), whereas leaf-and-spine traffic always crosses exactly two leaf uplinks plus a non-oversubscribed core. Any reported throughput gains likely reflect a comparison against an oversubscribed or suboptimally load-balanced baseline. _Network architects weighing fabric trade-offs for large deployments find peer analysis of hyperscaler papers on daily.dev._ ## Similar posts on daily.dev - [AWS Replaces Fat-Tree Data Center Networks with Random Graph Theory, Cutting Routers by 69%](https://daily.dev/posts/aws-replaces-fat-tree-data-center-networks-with-random-graph-theory-cutting-routers-by-69--rzssjbvtk) · InfoQ · 1 upvotes · 0 comments - [Switzerland built an alternative to BGP. Nobody noticed](https://daily.dev/posts/switzerland-built-an-alternative-to-bgp-nobody-noticed-9vzls8gls) · The Register · 0 upvotes · 0 comments - [Scale-across: Why the future of distributed AI isn’t in one data center](https://daily.dev/posts/scale-across-why-the-future-of-distributed-ai-isn-t-in-one-data-center-m2k1vyott) · Cisco · 0 upvotes · 0 comments - [10X Backbone: How Meta Is Scaling Backbone Connectivity for AI](https://daily.dev/posts/10x-backbone-how-meta-is-scaling-backbone-connectivity-for-ai-9yfrluyaf) · Facebook Engineering · 1 upvotes · 0 comments --- Tags: [#aws](https://daily.dev/tags/aws), [#networking](https://daily.dev/tags/networking) [View this post on daily.dev](https://daily.dev/posts/goodbye-leaf-and-spine-networks-ipspace-net-blog-qkpkkzvvc) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Goodbye, Leaf-and-Spine Networks? « ipSpace.net blog","url":"https://daily.dev/posts/goodbye-leaf-and-spine-networks-ipspace-net-blog-qkpkkzvvc","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/goodbye-leaf-and-spine-networks-ipspace-net-blog-qkpkkzvvc"},"datePublished":"2026-08-11T08:54:02.953Z","dateModified":"2026-10-14T02:35:16.269Z","description":"An AWS engineering paper claims a quasi-random graph (RNG) network fabric achieves 33% better throughput with 69% fewer routers, 27% cost savings, and 40%...","image":"https://media.daily.dev/image/upload/s--HRgLpUt6--/f_auto/v1722860399/public/Placeholder%2003","thumbnailUrl":"https://media.daily.dev/image/upload/s--HRgLpUt6--/f_auto/v1722860399/public/Placeholder%2003","isAccessibleForFree":true,"articleSection":"ipSpace.net Blog","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"ipSpace.net Blog","logo":"https://media.daily.dev/image/upload/s--xS7E82MT--/f_auto,q_auto/v1786438303/logos/ipspace?_a=BAMAMicg0","url":"https://daily.dev/sources/ipspace"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/goodbye-leaf-and-spine-networks-ipspace-net-blog-qkpkkzvvc","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"aws,networking","timeRequired":"PT8M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"ipSpace.net Blog","item":"https://daily.dev/sources/ipspace"},{"@type":"ListItem","position":3,"name":"Goodbye, Leaf-and-Spine Networks? « ipSpace.net blog"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/goodbye-leaf-and-spine-networks-ipspace-net-blog-qkpkkzvvc#faq","mainEntity":[{"@type":"Question","name":"What is the AWS quasi-random graph (RNG) network fabric and how does it differ from leaf-and-spine?","acceptedAnswer":{"@type":"Answer","text":"The AWS RNG fabric replaces spine switches with a prewired partial mesh device called a ShuffleBox, connecting leaf (ToR) switches directly via split uplink lanes. Traffic uses unequal-cost multipathing across multiple paths, including relay hops through other leaf switches. The design is mathematically a quasi-random expander graph, not truly random. It requires more ToR uplinks than a comparable leaf-and-spine fabric to achieve the same oversubscription ratio. Engineers evaluating hyperscale fabric alternatives track papers like this on daily.dev before committing to a design direction."}},{"@type":"Question","name":"Can a partial mesh or expander graph network fabric achieve higher throughput than a non-oversubscribed leaf-and-spine fabric?","acceptedAnswer":{"@type":"Answer","text":"No — in a true apples-to-apples comparison, a partial mesh cannot exceed the throughput of a non-oversubscribed leaf-and-spine fabric. Traffic in a partial mesh may traverse more than two leaf uplinks (using relay hops), whereas leaf-and-spine traffic always crosses exactly two leaf uplinks plus a non-oversubscribed core. Any reported throughput gains likely reflect a comparison against an oversubscribed or suboptimally load-balanced baseline. Network architects weighing fabric trade-offs for large deployments find peer analysis of hyperscaler papers on daily.dev."}}]} ```