Topics /
The most popular blog posts about systems & performance on Hacker News. 518 posts.
I started a reading group in 2021, a few months after joining Microsoft as a new grad on the Azure Databases team. The group was initially focused on database internals, which was my favorite subject at UW. Databases touch so many areas of CS: compiler construction in the query engine, memory management with the…
An RTL performance erratum in the Apple M3 Neural Engine throttles DRAM weight streaming throughput down to 17–19 GB/s from the nominal 45–60 GB/s, whenever the total weight size is an integer multiple of 1 MiB, which currently affects 7 of ANEMLL ’s 15 models. Avoiding the problematic path in the kernel DMA engine's…
Parallel evolution.
Performance results naturally depend on the underlying hardware. The following benchmarks show the execution times for sorting 50 million doubles using different sorting implementations. The measurements were taken on an Apple M1 system using Clang and on an AMD Ryzen 3 Linux system using GCC , both compiled with the…
Microsoft’s in-house Linux, the distribution that grew out of CBL-Mariner, just hit public preview as a general-purpose cloud OS you can run on any Azure VM and coming soon to WSL.
zswap and zram are fundamentally different approaches with different philosophies. If in doubt, use zswap.
In the prequel, we introduced Neptune, a Direct3D protocol forwarding layer for VirtIO. Neptune allowed us to serialize Direct3D API calls across the hypervisor boundary and this allowed us to run Wine games on a Linux guest with a Linux host faster than with DXVK directly in the guest. Admittedly, the payoff there…
We have seen in a previous post that switch statements are more complex than they seem. The compiler is able to implement them as a decision tree (even though maybe it shouldn't). Is that everything it can do?
std::simd Can't Express 90% of Real SIMD Code
zeroserve is a high-performance HTTPS server that runs eBPF scripts in userspace ( intro ). Now it's got a Caddy-compat mode - when provided a Caddyfile, zeroserve JIT-compiles it to eBPF and then to native x86_64/ARM64 machine code, and runs it in an io_uring event loop.
This document describes how Fil-C supports longjmp , setjmp , setcontext , getcontext , makecontext , and swapcontext in a totally memory-safe way. In particular, no misuse of those APIs in Fil-C can lead to stack corruption or any other violation of Fil-C's capability model.
Why compaction is needed for large language models and how Pi implements it
You're probably reading this page because you've attempted to access some part of my blog (Wandering Thoughts) or CSpace , the wiki thing it's part of. Unfortunately you're using a browser version that my anti-crawler precautions consider suspicious, most often because it's too old (most often this applies to versions…
I've seen lots of chatter about Fil-C recently, which pitches itself as a memory safe implementation of C/C++. You can read the gritty details of how this is achieved, but for people coming across it for the first time, I think there is value in showing a simplified version, as once you've understood the simplified…
Diskless Linux Boot
My personal website and blog
How I obtained my own AS number and IPv6 prefix, set up a FreeBSD BGP router with FRR, and built a tunnel overlay to bring globally routable addresses to servers that already have provider-assigned...
My homepage now runs on OpenBSD with httpd(8) and Let's Encrypt via acme-client(1) . I previously hosted it on GitHub Pages and then eventually moved to Cloudflare Pages because both options were free and easy. Free and easy is cool, and I understand that writing software full-time leaves us wanting absolutely nothing…
In the previous post, I mentioned that buildcache has some unique properties compared to ccache and sccache. One of them is its Lua plugin system, which lets you write custom wrappers for programs that aren’t compilers in the traditional sense. With Bug 2027655 now merged, we can use this to cache Firefox’s WebIDL…
Disclaimer: An earlier version of this post claimed the structure is wait-free, this is incorrect. Being wait-free requires that failure or suspension of any thread can’t cause failure or suspension of another thread. This queue in fact does not fulfill that requirement. The main section which discusses the wait…
Buy an RTX 5090, the fastest consumer GPU money can buy, and you get 104.8 TFLOPS of FP32 compute. Ask it to do double-precision math and you get 1.64 TFLOPS. That 64:1 gap is not a technology limitation. For fifteen years, the FP64:FP32 ratio has been slowly getting wider on consumer GPUs, widening the divide between…
Proofcraft’s Chief Engineer Rafal Kolanski has completed a major “architecture split” overhaul of the seL4 proofs, resulting in the removal of 250k lines of proof and strengthening the link between proofs on different architectures. These changes will significantly reduce future effort required to maintain and extend…
Zone setting changes used to rebuild entire DNS zones, even ones with tens of millions of records. Here is how we built a planner that picks the smallest safe build, cutting full builds by 40%, peak memory by 85%, and halving build delay.
Justin Cormack, who worked on MirageOS back in the day, called me at the pub to talk about why unikernels are coming back. The friction that made them hard is gone, the operating system is design debt, and if you care about security you should seriously consider them.
How customer needs shaped Oxide's Kubernetes integrations.
For some reason, nobody's written about what it takes to crawl a big chunk of the web in a while: the last point of reference I saw was Michael Nielsen's post from 2012 .
I spent a whole month researching FreeBSD's kernel's virtual memory system to understand why tools like htop, btop and fastfetch report different information about RAM usage. I ended up submitting patches to all of these projects.
