Ratul Mahajan

@vishalmisra @anand_raj That’s a great Modelo commercial!

Dream fulfilled! With the master blaster himself - Sir Isaac Vivian Alexander Richards!

The Times did a nice piece on @om: nytimes.com/2026/06/26/tec…

@vishalmisra @om He would have loved that :(

@ratulm @om :-(. We used to have these extended and excited discussions about T20 cricket and I was going to invite him to come see a Unicorns game this weekend in Oakland!

@om passed away yesterday. I'll remember him as a dear friend, a fellow coffee lover, a most original voice in tech writing, and a mentor who always shared his opinion but never enforced it. I love this picture of him.

@vishalmisra @om Heart issues. He was on the transplant list but one didn't become available in time.

@ratulm @om Oh no - that’s so sad! What happened?

Rituals are silly, but fun too. Here are @arvind_uw and I hooding our PhD student, @xiangfeng_zhu. His thesis showed how to design and implement networks that are hyper-customized to applications' needs rather than requiring applications to work around whatever the network stack happens to provide. He is now off to help machines think at Thinking Machines. Good luck, Xiangfeng!

Whoa -- I am so lost writehuman.ai

@JeffDean @uwcse @MBalazinska Thank you for sharing your story and the lessons with our graduates. The contrast between the constant movement of your life growing up and the stability of your professional life is fascinating.

@uwcse @MBalazinska Here are a few of the pieces of advice that I shared:

I enjoyed giving the commencement address at the University of Washington Allen School @uwcse graduation this evening. So many happy students and their families and friends! Congratulations to all the graduates of the class of 2026! 🎓 Thanks for inviting me, @MBalazinska!

This feature make no sense. With Replit ("Turn ideas into apps in minutes — no coding needed"), shouldn't the CEO be able to build the app themself?

We wrapped up the LLM systems reliability seminar last week and learned about a lot of cool ideas and tricks. What struck me most though: many ideas felt not relevant to current practice despite being only 3-4 years old. A short summary deck: #slide=id.p" target="_blank" rel="nofollow noopener">docs.google.com/presentation/d…

James Hamilton is who infra engineers want to be when they grow up. So great to see RNG getting a shout out from him. perspectives.mvdirona.com/2026/06/flat-d…

new paper explainer! a network problem sat unsolved for a decade because the fix needed 80× more memory than the hardware has. the solve is running in prod now! arXiv link ↓

Amazon unveiled “Resilient Network Graphs,” (RNG) a data center network that reduces hardware needs by 69% and raises throughput by 33%. It is now default for most AWS workloads. They revealed that it has been quietly deploying the design across its data centers since last year, and it is now the default data center network for most AWS workloads. It replaced tree-shaped datacenter networks with flatter random ones that waste less capacity. For decades, fat-tree networks worked because they were predictable, but their layered shape can concentrate traffic at choke points while other links sit underused. So the problem is that fat-tree networks are easy to run, but their hierarchy can trap traffic on a few links while other links sit unused. “Resilient Network Graphs,” (RNG) fixes this by connecting routers in a flat quasi-random graph, so many different paths exist between servers instead of a few fixed routes through upper layers. RNG attacks the problem by flattening the fabric into a quasi-random graph, where many small independent paths replace a few privileged routes. Its routing system, Spraypoint, spreads traffic across many separate paths, while its ShuffleBox cabling device makes the random-looking wiring practical to build and expand. Instead of asking every packet to chase the shortest path, Spraypoint fans traffic outward and then guides it back through distributed waypoints, creating many edge-disjoint paths without requiring exotic switch memory. The authors tested RNG in 2 real Amazon production fabrics and compared it with fat-tree networks using transport and storage workloads. The main result is that RNG matched fat-tree application performance, found far more separate paths than common routing methods, and was estimated to cost 9% to 45% less. The hard part is not the idea, but the engineering, because routing in a random mesh needs smarter path selection and the physical system must manage millions of fiber connections without becoming impossible to operate. This is important for AI clusters because training traffic is huge, synchronized, and sensitive to congestion, so a network that spreads load better can make expensive GPUs spend less time waiting. ---- Link – arxiv. org/abs/2604.15261 Title: "RNG: Flat Datacenter Networks at Scale"

@BenjDicken Thanks for the shoutout. Giacomo is here as @mino98; he prefers to just lurk. (Sorry that I outed you, Giacomo.)

Combining deep math theory with practical implementation leads to many cool things. This is the heart-and-soul of computer science and engineering. AWSs latest blog is a beautiful example, describing and visualizing the transformation from tree-based networks to pseudo-random graphs in their DCs. Props to @ratulm and the other authors (who need to get on X dot com ASAP)

Most people thought flat data center networks would never work at hyperscale. The @awscloud team figured it out. Resilient Network Graphs are a completely new network architecture, now live. 33% better throughput, 40% less network power. aboutamazon.com/stories/aws-ra…

A passive optical device called a ShuffleBox and a routing protocol called Spraypoint just made flat datacenter networks as practical as fat trees, with 69% fewer routers and up to 33% better throughput. amzn.to/4uIdylW

The real story: arxiv.org/abs/2604.15261

Look, Ma, no more trees! wired.com/story/amazon-t…