August Update

Sam Newman

— 3 min read
Share

The summer weather is just about holding on here in the UK, but more autumnal weather is clearly on the way, and with it a very busy schedule for me from September till the end of the year.

This month, amid the now constant hum of AI hype, I've actually managed to make some decent progress on the book!

Book Update

The chapter on progressive collapse is just waiting on me to check review comments before it gets released (unleashed?) on early access. It's been a really fun chapter to write, partly because of the research I did for it. I got to read (well, skim) reports from NIST. Good times.

I also found time to revisit the earlier chapters to update them to highlight a few patterns. The chapter on Retries got a lot of changes so I've asked for another reviewer pass before I push that to early access - I significantly beefed up the discussion around idempotency to the point where it might end up being a chapter in its own right, although I'm waiting to hear from the reviewers about their thoughts before I make that call.

Things I Have Found

  • Inside the relentless race for AI capacity. I am a sucker for highly visual story telling. I can struggle with large blocks of text, which is largely why my own books feature a lot of diagrams. This article from the FT on the data centres being built to support the AI hype train therefore is exactly my sort of thing. Come for the pretty pictures, stay for the content. And unlike a lot of the FT content, no sub required.
  • Rethinking Distributed Computing for the AI Age. This fascinating ACM article makes the argument that how we think about distributed computing is fundamentally based on the assumption that problems are trivially parallelizable, but that AI workloads fundamentally are not. The fact that the current architectures require the synchronization of more data between nodes is problematic, especially if we take author Akshay Mittal at his word when he states that: "I’ve observed that data movement often consumes orders of magnitude more energy than computation itself.". So a rallying call to think differently about both our software architecture and our infrastructure.
  • ClusterMax rating. Yet another piece in the space of AI infrastructure, this analyst rating model looks at the various GPU cloud vendors. I found this after seeing Dylan Patel from SemiAnalysis delivery the closing presentation from the recent Superstream I hosted for O'Reilly. What was enlightening for me was the huge growth in the neocloud space, and especially the fact that many of these new upstarts' biggest clients are actually some of the existing incumbent hyperscalers.
  • A better diff. A bit of a gear change this, with GitClear making the case that a better diff algorithm would make code reviews faster. The normal Meyers diff algorithm we use dates from 1986, and this research argues it does a bad job of dealing with code edits that are effective no-ops, such as simple movement of code. Now, a pinch of salt is needed here as GitClear are promoting their own commercial tool with this research, but nonetheless it does raise the prospect for other folks to come up with their own alternatives.

Forthcoming Events

I said the last few months of the year were going to be busy for me, and I wasn't lying:

  • GOTO Copenhagen, September 29th - October 3rd. Not many spots left for the two public workshops I am doing at this event. I'll be running a one day workshop on microservice communication, and another on resilient distributed systems.
  • API Days India, October 8th - 9th. As well as a talk, I will be delivering a one day workshop on October 10th on resilient distributed systems (details on that will go live soon)
  • Techorama NL, October 27th - 29th. Another talk and a chance to catch my workshop on resilient distributed systems.
  • Trifork Masterclass, Amsterdam, October 30th - 31st. This will be the last chance to attend my public two day microservices workshop in 2025.
  • LeadDev Berlin, November 3rd - 4th. The schedule isn't yet live, but I will be delivering a talk on the nature of resiliency.
  • MQ Summit, November 5th - 6th. I'll be delivering a keynote on the nature of meaning, and specifically what asynchronous actually means.
  • Architecture Conference Tokyo, November 20th - 21st. What's better than one keynote from me? How about two! Really looking forward to what will be my first trip to Japan. I will also be delivering a workshop while in Tokyo, details TBC.

That's all for now. By next week, I'm sure all this AI noise will have diminished, and we'll be back to writing programs by carving runes into stone tablets, just like our forefathers.

Bye!

Read more