New blog post, reflecting on nearly seven years since the Firecracker launch, and how we're using Firecracker to power serverless databases (in Aurora DSQL) and infrastructure for AI agents (in Bedrock AgentCore). Here's the post: brooker.co.za/blog/2025/09...

NULL BITMAP: Using Huffman Coding for Sampling Discrete Distributions: buttondown.com/jaffray/arch...

Seattle friends - in case you missed it: lu.ma/umtmqb38 The Seattle Systems group is hosting their very first meetup! @marcbrooker.bsky.social will be presenting Aurora DSQL. Excited to see more systems groups popping up (thanks @eatonphil.bsky.social). My colleague @devhawk.net is co-organizing.

People often ask me about the differences in architecture between Amazon Dynamo (the 2007 SOSP paper), DynamoDB (the AWS serverless NoSQL database), and Aurora DSQL (the AWS serverless SQL databases). I memoized the response on my blog. brooker.co.za/blog/2025/08...

If you're into databases and you're in Seattle, @marcbrooker.bsky.social is talking at Seattle Systems on Aug 21: lu.ma/umtmqb38

LLMs are more powerful, more dependable, more efficient, and more flexible when deployed as a component of a carefully designed system. New blog post: brooker.co.za/blog/2025/08...

I had a great time talking to Ben and Will from Antithesis on the BugBash podcast about software testing and validation at AWS. In this wide-ranging hour we talk deterministic simulation testing, fuzzing, formal methods, the history and future of testing, and much more.

Fun conversation with @jkxosound.com about where I think things are going in the AI agents space, Strands Agents, and rats 🐀

Prototyping agents is easy, fast, and fun (especially if you use Strands). But how to you get them into production with all the scalability, security, and resilience you need? We've been building Bedrock AgentCore to answer exactly that question. Check it out: aws.amazon.com/blogs/aws/in...

I sat down with Darko Mesaros to talk about Kiro, specification-driven development, and putting AI-powered development in context of the history of computing. Is an LLM a compiler? What is a spec? How are specs like SQL? Check it out: www.youtube.com/watch?v=OdD9...

I've been using Kiro as my primary IDE for a couple months, and am super happy to see it launch. Spec-driven development is super powerful. More reliable, more repeatable, and easier to control than prompt-based vibe coding. Kiro's hooks are also pretty great. I use them to keep clippy happy :)

Finally got to fixing the "related posts" feature in my blog. The Amazon Titan embedding model on Bedrock, plus cosine similarity, does a nice job:

🚀 New AWS Bites episode! We tried Amazon #DSQL so you don’t have to (but you might want to) 👀 Luciano and Eoin put Aurora DSQL through its paces in #AWS Bites #Podcast Episode 145 🧵 Here’s what we found:

New blog post, with some career advice: brooker.co.za/blog/2025/06...

Here's the answer. On the left, we see the great "exponential benefit at linear cost" impact of redundancy. But that quickly tapers off as the single point of failure dominates. What can this teach us about building good HA architectures?

Wrote a monte carlo simulator in Python. Took 300 seconds to run. Asked Q CLI to translate it to Rust, compile, and run. Took 48 seconds, including the translation, and gave exactly the same result. Is Claude 4 the best Python compiler?

Availability intuition quiz! I have an architecture with N redundant components, each available 99% of time time, and one non-redundant component available 99.999% of the time. Sketch, on the provided axes, the end-to-end availability curve versus N.

New blog post, about some of my favorite papers from this year's HotOS program. Love Little's law? Hate fork? Think static analysis of shell programs sounds fun? Had enough of the f threshold model? Check it out for some papers to read: brooker.co.za/blog/2025/06...

"You will never experience eventuality. Eventual consistency is an abstract concept, not a guarantee you can count on."

Interested in how we test systems at Amazon? In a new article in Communications of the ACM, Ankush and I write about the evolution of systems correctness and testing practices at AWS. We cover classic and formal approaches, and new ones like deterministic simulation. cacm.acm.org/practice/sys...

Aurora DSQL is now Generally Available! I couldn't be more proud of the team that brought DSQL to life, and excited about how our customers are going to build on it. 🧵

In a new guest post from #AWS Sr. Principal Engineers @nikomatsakis.com and Mark Bowes take us inside Aurora DSQL's development from scaling write operations without two-phase commit, to overcoming garbage collection hurdles, and embracing Rust. www.allthingsdistributed.com/2025/05/just...

Last week I had the privilege to give a keynote at the International Conference on Performance Evaluation (ICPE'25). It wasn't recorded, so I captured the key points as a blog post: brooker.co.za/blog/2025/05...

Recently, one of our testing tools at AWS found an issue with transaction visibility in PostgreSQL, where the order that transactions become visible differs between the primary and replicas. Sergey Melnik dives deep on the AWS Database Blog: aws.amazon.com/blogs/databa...

I enjoyed @alexmillerdb.bsky.social's "Decomposing Transaction Systems" blog so much I had to write up how I'd decompose Aurora DSQL in a new blog post: brooker.co.za/blog/2025/04...

This is an excellent insight into the nature of transaction systems. Worth reading if you have any interest in databases.

I generally like Mitutoyo products, but this travel comb sucks.

Exciting to see another great step forward in the Rust standard library verification contest with the addition of VeriFast. www.linkedin.com/posts/r%C3%A...

@marcbrooker.bsky.social breaks down how they've architected a fully ACID-compliant database service that combines simple, serverless management with high availability and massive scale on AWS Aurora DSQL. youtube.com/shorts/dScUi...

It's widely known that sharing a queue across multiple servers, rather than queue-per-server, often helps reduce latency and improve utilization. But when is one queue better? In my new blog post, I look at one case: different classes of work. Read it here: brooker.co.za/blog/2025/03...

I had a great time chatting to John Kutay on the What's New In Data Podcast about Aurora DSQL, the architectural choices we made, and the trade-offs in the database space. Check out our conversation here: www.youtube.com/watch?v=Q3CY...

What Justin is talking about here is close to how Aurora DSQL's internal storage works. Even in our case (building SQL on top of the storage), pure K/V leaves a lot of optimization opportunities on the ground.

Love the explanation of achieving coordination-free local properties, ie, avoiding coordination, while still providing useful isolation levels at scale and across regions.

Wrote a blog post reflecting my thoughts on DeepSeek, NSF funding and system research communities in general. Apologies for the bold claims -- hope they can invite some discussions. blog.xiangpeng.systems/posts/system...

One challenge of handling partial/gray failures in distributed systems is telling 'healthy' from 'unhealthy'. Even in terms of error rate, it can take a surprisingly high number of samples to differentiate between normal and abnormal hosts. Here's what it looks like:

Bizarre.

One aspect of the work we describe in Systems Correctness Practices At AWS (dl.acm.org/doi/10.1145/...) is the work Ankush and team have been doing on closing the loop between formal specification and implementation of distributed systems.

In this month's ACM Queue, @ankushpd and I write about some of the methods and tools we apply to systems correctness at AWS: from testing, to simulation, to fault injection, to formal proofs. Read it here: dl.acm.org/doi/10.1145/...

Continuing from yesterday's blog post, today I look at an interesting isolation edge-case from the literature: Fekete's Anomaly. How does it fit in to our theme of achieving performance, scalability, and reliability through reducing coordination? New blog post here: brooker.co.za/blog/2025/02...

Excited to share our ICDT 2025 paper on a theoretical framework to connect CRDTs and the CALM Theorem! We formalize the intuitions given in the Keep CALM and CRDT On work about the gap between CRDT guarantees and monotonic queries. arxiv.org/pdf/2502.00222

New blog post, on the role of versioning in avoiding coordination in distributed systems: brooker.co.za/blog/2025/02...

A while ago @michal.ostruszka.me shared a talk by @marcbrooker.bsky.social on resiliency strategies used in the AWS SDK. Some weeks passed, and now we have an implementation of "adaptive retries" available as one of the retry patterns in Ox, thanks to Kamil Lontkowski.

I had a great time talking to Kaivalya Apte about Aurora DSQL on the Geek Narrator podcast. Watch our episode here: www.youtube.com/watch?v=ONkf... We cover the design of Aurora DSQL, interesting aspects of the design, PostgreSQL, Firecracker, and trade-offs of database systems. A lot in an hour!

Amazon MemoryDB multi-region is now generally available. If you're a dist sys geek, you may be interested to know that this multi-region active-active eventually consistent database is based on conflict-free replicated data types (CRDTs).

The most valuable thing you can do for your career is to be thoughtful about how you spend your time. One of the tools I use most to do that, Mai-Lan Tomsen Bukovec's principal roles framework, is now available outside Amazon for the first time.

New blog post, diving a little deeper on Snapshot Isolation and Serializability, through the lens of the coordination required to implement them: brooker.co.za/blog/2024/12...

My third talk at re:Invent this year was about patterns for system resilience. Here, I talk about the trouble with retries, metastability, the power of simulation, and some techniques for tackling tail latency: www.youtube.com/watch?v=rvHd...

This is one of my favorite talks of #reinvent 2024, certainly one of my favorite styles of talks where it's the theory and algorithms behind distributed and scaled systems. Another one from @marcbrooker.bsky.social www.youtube.com/watch?v=rvHd...

My second re:Invent talk is up! In this talk, I dive into the internals of thew new Aurora DSQL, looking at how we achieve each of the ACID properties, and highlight some of the decisions we made as we designed the product. www.youtube.com/watch?v=huGm...

"While database researchers often focus on database engine architectures, Andy argued that surrounding topics, such as network connection handling (e.g., database proxies), receive little attention despite their practical importance." Strong agree.