Jan de Mooij writing on the SpiderMonkey blog:

In September, machine learning engineers at Mozilla filed a bug report indicating that Firefox was consuming excessive memory and CPU resources while running Microsoft’s ONNX Runtime (a machine learning library) compiled to WebAssembly.

The issue with the ONNX module was that the Ion compiler backend took a long time and used a lot of memory to compile it. On my Linux x64 machine, Ion-compiling this module took about 5 minutes and used more than 4 GB of memory.

⋮

With these changes, Ion can compile the ONNX Wasm module in less than 3.9 seconds on my machine, more than 75x faster than before these changes.

I find it satisfying reading about optimisations like this. The same end-state is achieved, but in vastly less time and space—it feels like magic.

Kevin Guthrie writing on the Cloudflare blog:

This small and pleasantly-readable function consumes more than 1.7% of pingora-origin’s total cpu time. To put that in perspective, the total cpu time consumed by pingora-origin is 40,000 compute-seconds per second. You can think of this as 40,000 saturated CPU cores fully dedicated to running pingora-origin. Of those 40,000, 1.7% (680) are only dedicated to evaluating clear_internal_headers. The function’s heavy usage and simplicity make it seem like a great place to start optimizing.

The scale of Cloudflare’s operation is staggering when put in terms like this.

Josh Aas writing on the Prossimo blog:

The River reverse proxy recently has come a long way since we announced the project in February.

River is a project of Prossimo, itself a project of the Internet Security Research Group (ISRG) project. ISRG are the folks behind Let’s Encrypt. The River reverse proxy is implemented in Rust.

Just about every significantly-sized deployment on the Internet makes use of reverse proxy software, and the most commonly deployed reverse proxy software is not memory safe. This means that most deployments have millions of lines of C and C++ handling incoming traffic at the edges of their networks, a risk that needs to be addressed if we are to have greater confidence in the security of the Internet.

Our own goal is to have River ready to replace Nginx and other reverse proxy software used by Let’s Encrypt within the next year, and we encourage other organizations to start considering where they might start to improve the security of their networks with memory safe proxy software.

That would constitute quite a significant production deployment for the project.

Loris Cro:

In front of you are two personal websites, each used as a blog and to display basic contact info of the owner:

1. One is a complex CMS written in PHP that requires a web server, multiple workers, a Redis cache, and a SQL database. The site also has a big frontend component that loads as a Single Page Application and then performs navigation by requesting the content in JSON form, which then gets “rehydrated” client-side.
2. The other is a collection of static HTML files and one or two CSS files. No JavaScript anywhere.

If you didn’t know any better, you would expect almost all normal users to have [2] and professional engineers to have something like [1], but it’s actually the inverse: only few professional software engineers can “afford” to have the second option as their personal website, and almost all normal users are stuck with overcomplicated solutions.

For №2 to be successful for “normal” users I think there needs to be an approachable UI to be able to make changes on the fly… and the implementation gets complicated quickly if you’re doing it on the website itself.

The funny thing is that we we had such a thing in the early days of the web: native apps for authoring a website and uploading it, such as Frontpage, Dreamweaver, or Netscape Composer. It’s strange that these seem to have fallen out of favour now.

Julio Merino:

All of this growth has been in service of ever-growing programs. But… even if programs are now more sophisticated than they were before, do they all really require access to a 64-bit address space? Has the growth from 8 to 64 bits been a net positive in performance terms?

Let’s try to answer those questions to find some very surprising answers.

⋮

We observe massive differences in the machine code generated for the trivial main function. The 64-bit code is definitely smaller than the 32-bit code, contrary to my expectations. But the code is also very different; so different, in fact, that ILP32 vs. LP64 doesn’t explain it.

Was the move to 64-bit CPUs all upside? Julio runs the numbers to show some pros and cons.

Eugen Rochko writing on the Mastodon blog:

Mastodon 4.3 just landed! If you’re a mastodon.social user, you might have already seen some of this in action as we’ve been gradually rolling out these updates over the course of the last 11 months in nightly releases, but we’re finally making a new stable release available to the community. If you use a different server, you will get access to these improvements once your server operator upgrades.

This one’s been in the works for a long time. The notification improvements are most welcome. I have been using the Phanpy Mastodon client for a while, partly due to grouped notifications. Glad this has functionality has made it to the official client now.

sdomi:

In 1999, some members from the MMC Association decided to split and create SD Association. But nobody seems to exactly know why.

I’m sure there’s some folks out there that were part of these associations. Hopefully they will get in touch with sdomi and enlighten us all.

Nathan Dyer:

A couple weeks ago I decided to purchase a MacBook Air M2, with the specific purpose of running Asahi Fedora Remix. I chose this particular model because it’s the newest chipset that is supported by Asahi, and Best Buy was having a $200 off sale on them, presumably before Apple announces a new slate of laptops and stops offering the aging M2-based systems.

There were several things I really loved about the system (battery life was astounding, the trackpad was marvellous, and the display was beautiful), but ultimately I made the decision to return it; the experience just wasn’t what I needed for a daily driver.

Seems like there’s still a bit of polishing required for Asahi to match the experience of Linux on non-Apple hardware.

Arnaud Carré:

The Commodore Amiga 500 had a blitter chip. Its main function was to move bitmap graphics from one location to another while applying logical operations. The Amiga’s blitter could handle up to three bitmap sources at once and perform logical operations between them. To specify which operation to use, you needed to set an 8-bit value in the chip, known as the “minterm.”

Three bitmap sources and an 8-bit value to control logical combinations! Doesn’t that sound like a primitive version of the modern AVX vpternlogd instruction?

An unexpected correlation between a particular AVX-512 instruction and 35 year old Amiga graphics hardware.