Last week I had the pleasure of attending RustForge in Wellington, New Zealand. I gave a talk titled “Wild performance tricks”. You can watch a recording of my talk . If you’d prefer to read rather than watch, the rest of this post will cover more or less the same material. The talk shows some linker benchmarks, which…
Have you ever asked yourself what if we could replace any message broker with a very simple one using only two UNIX signals? Well, I'm not surprised if you
I've always been told that caching is a tool to make software faster. That, given some careful considerations to consistency, caching makes it so that when...
No-one really likes engineering war stories, but this one's relevant because there's a moral to it. I've talked before about defining technical debt as technical decisions that provide immediate value, but with long-term negative impact if they aren't cleaned up. Sometimes introducing technical debt
Writing a Hypervisor in 1,000 Lines
Recently a colleague mentioned building NixOS images that run unchanged on multiple architectures. Given the past adventures on this blog with systemd-repart and cross-compiling NixOS, I decide to give this a go.
To start off, we can leverage the STL for this calculation. std::count_if() more precisely:
AWS Senior Principal Engineers, Niko Matsakis and Marc Bowes, take us inside Aurora DSQL's development: scaling write operations without two-phase commit, overcoming garbage collection hurdles, and embracing Rust for both data and control planes.
In a service with capacity $\mu$ requests per second and background load $\lambda_0$, the usable headroom is $H = \mu - \lambda_0 > 0$. When $M$ clients align, e.g., after a cache expiry, at a cron boundary, or as a service returns from an outage, the bucketed arrival rate can
There is oft-quoted advice that Kafka does poorly as a job queue. I’ve experienced this myself, and I wanted to formalize it a bit.
Putting FreeBSD’s “power to serve” motto to the test. On Thanksgiving morning, I woke up to one of my web services being unavailable. All HTTP requests failed with a “503 Service unavailable” error. I logged into the console, saw a simplistic “Runtime version: Error” message, and was not able to diagnose the problem…
Exponential backoff with jitter is de rigeur for making service calls. This code, or something like it, probably looks really familiar: func do(ctx context.Context) error { const ( maxAttempts = 10 baseDelay = 1 * time.Second maxDelay = 60 * time.Second ) delay := baseDelay for attempt := range maxAttempts { err :=…
Every transactional system must execute, order, validate, and persist transactions.
So how hard could it be. As input we have something like Fri, 17 Jan 2025 06:07:07 in UTC, and we’d like to turn this into 1737094027, the notional (but not actual) number of seconds that have passed since 1970-01-01 00:00:00 UTC. Trying to figure this out led me to discover many ‘surprise features’ and otherwise…
A few weeks back I got a note on Bluesky, linking me to a website offering a free computer. The owner was in Cambridge, not far from me, and the computer was an HP 9000/300 Series - a Model 340, specifically.
Model minimalistically Start from a tiny core, and always keep a working model as you extend. Your default should be omission. Add a compone...
Windows 98 runs surprisingly well in QEMU via UTM SE, but it requires some care in setting it up. It’s a great way to run old 90s Windows and DOS software on your iPad (and Mac too, though you have… Continue reading →
Write your first hypervisor from scratch, in 1K LoC.
For the last few years, I've been developing caching library, and today I'd like to talk about the evolution of caches in Go and where we stand today. Especially since Go 1.24 was officially supposed to focus on caching improvements, but I haven't heard much news about them - time to fix that .
How hot code loading solves client-specific engineering problems in my startup Alzo
Sending a ping by creating an ICMP socket normally requires root: you can’t create a raw socket to send ICMP packets without it. The ping command line tool works without root however, how is that possible? It turns out you can create a UDP socket with a protocol flag, which allows you to send the ping rootless. I…
See how AWS Lambda arm64 stacks up against x86_64 in real-world benchmarks across CPU, memory, and I/O workloads using Node.js, Python, and Rust.
Most C++ and Rust thread pools run 10x slower than OpenMP on fork-join workloads. Fork Union closes the gap to 20% with 300 lines, dodging mutexes and CAS.
Let's explore and deobfuscate the input stack on Linux. Our aim is to understand its components and what each does. Input handling can be divided into two parts, separated by a common layer. We’ll try to make sense of all this, one thing at a time, with a logical and coherent approach.
To @AndrewHart: “Our app is shit and we know it’s shit”. My life changed forever when a senior exec from one of the world’s largest retailers messaged me on Twitter. This was back in 2017. My outdoor AR navigation demos had just gone viral, and my new open-source project for...
System76 built their first workstation-class Arm PC, the Thelio Astra, and it's marketed for streamlined autonomous vehicle development. But I'm not an automotive developer, just someone who enjoys Linux, Arm, and computing. So I was excited to spend a few weeks (which turned into a few months) testing the latest…
My CppCon 2025 talk, Practical Reflection, is now online, and goes over this example in more detail and more digressions. Check it out!
In this blogpost I will step by step show you how to implement a highly efficient transpose kernel for the Hopper architecture using Mojo. The best kernel a...
