Linus Torvalds Talks About Rust Adoption and AI (zdnet.com) 48
"At The Linux Foundation's Open Source Summit China conference, Linus Torvalds and his buddy Dirk Hohndel, Verizon's Head of the Open Source Program Office, once more chatted about Linux development and related issues," reports ZDNet:
Torvalds: "Later this year, we will have the 20th anniversary of the real-time Linux project. This is a project that literally started 20 years ago, and the people involved are finally at that point where they feel like it is done... well, almost done. They're still tweaking the last things, but they hope it will soon be ready to be completely merged in the upstream kernel this year... You'd think that all the basics would have been fixed long ago, but they're not. We're still dealing with basic issues such as memory management...."
Switching to a more modern topic, the introduction of the Rust language into Linux, Torvalds is disappointed that its adoption isn't going faster. "I was expecting updates to be faster, but part of the problem is that old-time kernel developers are used to C and don't know Rust. They're not exactly excited about having to learn a new language that is, in some respects, very different. So there's been some pushback on Rust." On top of that, Torvalds commented, "Another reason has been the Rust infrastructure itself has not been super stable...."
The pair then moved on to the hottest of modern tech topics: AI. While Torvalds is skeptical about the current AI hype, he is hopeful that AI tools could eventually aid in code review and bug detection. In the meantime, though, Torvalds is happy about AI's side effects. For example, he said, "When AI came in, it was wonderful, because Nvidia got much more involved in the kernel. Nvidia went from being on my list of companies who are not good to my list of companies who are doing really good work."
Switching to a more modern topic, the introduction of the Rust language into Linux, Torvalds is disappointed that its adoption isn't going faster. "I was expecting updates to be faster, but part of the problem is that old-time kernel developers are used to C and don't know Rust. They're not exactly excited about having to learn a new language that is, in some respects, very different. So there's been some pushback on Rust." On top of that, Torvalds commented, "Another reason has been the Rust infrastructure itself has not been super stable...."
The pair then moved on to the hottest of modern tech topics: AI. While Torvalds is skeptical about the current AI hype, he is hopeful that AI tools could eventually aid in code review and bug detection. In the meantime, though, Torvalds is happy about AI's side effects. For example, he said, "When AI came in, it was wonderful, because Nvidia got much more involved in the kernel. Nvidia went from being on my list of companies who are not good to my list of companies who are doing really good work."
The GNU/Hurd team seyz (Score:2)
Re: (Score:2)
I have no problem with long-winded project. Most good software I use every day has taken years or decades to mature. Hell, Linux itself is over 30 years old.
I've always been skeptical about the stupid "Move fast and break things" mantra: the tech bros who keep repeating it think they're clever, but those of us who have been around the block know all it really creates is trail of broken software.
Re: (Score:2)
I'd argue there's a significant difference between those long-lived projects you might rely on (Linux, Gimp, etc.) as opposed to projects like Hurd. Linux may be 30+ years old, for example, but most of that 30 years has been spent on constant iteration and improvement of an already very usable product. Hurd is more in the "Duke Nukem Forever" class... although arguably that's being a tad unfair to the developers of DNF - since they did eventually release a product.
Recycled (Score:3)
https://m.slashdot.org/story/432284
Re: (Score:2)
And I thought you were going to be pointing at https://linux.slashdot.org/story/24/08/23/1857221/linux-creator-torvalds-says-rust-adoption-in-kernel-lags-expectations [slashdot.org], although this new version covers a wider range of topics.
Re: Recycled (Score:2)
Re: (Score:2)
I had to replace my mouse earlier and ended up having to retype (and mistype) your link, no idea what I got wrong.
Re: Realtime Linux? (Score:3)
QNX was designed to be real-time from the start. Linux was not. As you say, half assing real-time can be disastrous, so moving slowly is good.
Re: (Score:3)
Indeed. 20 years for retrofitting something like that is really not surprising and needed if you want to do it right. On the plus side, when it is done and stable, it is a capability with a massive deployment footprint.
Re: (Score:2)
We used QNX in our industrial machinery in the mid-90s.
QNX uses a microkernel.
That makes real-time easier, but you pay the price in performance when you move stuff like networking and file systems to user processes.
Re: Realtime Linux? (Score:2)
So, professor Tannenbaum has been right all along? They should have worked on a realtime version of Minix, instead of Linux. Or even GNU Hurd.
Re: (Score:3)
So, professor Tannenbaum has been right all along? They should have worked on a realtime version of Minix, instead of Linux. Or even GNU Hurd.
Performance vs. reliability and security has been an ongoing debate since early times. IRIX 5 included xhost + in the default X config for convenience. Windows 4.0 merged the User and GDI memory spaces for graphics performance. Early GUI operating systems like AmigaOS had no memory protection, which let you make powerful modifications to running programs but also precluded having any security.
There is not one correct answer, but for servers, microkernel arguably does make more sense. Security is more import
Re: (Score:2)
Security makes sense for most servers nowadays.
But I doubt most servers need "realtime".
No one said otherwise. You're arguing with yourself. The benefit of a microkernel for servers is security, not realtime. Microkernel OSes don't have to be realtime, they just make realtime easier (but at the expense of performance.)
Re: Realtime Linux? (Score:5, Insightful)
So, professor Tannenbaum has been right all along?
No, not really.
Microkernels have advantages, and real-time is one of them.
But they also have drawbacks, like networking and file system performance, which matter more to most people most of the time.
For a desktop or server OS, monolithic is generally better and has won in the marketplace.
Re: (Score:2)
That makes real-time easier, but you pay the price in performance when you move stuff like networking and file systems to user processes.
Actually moving networking out of the kernel may provide significant performance gains - read up on the DPDK project.
Re: (Score:2)
Re: (Score:2)
I can get ~10gbps out of a mid-level Xeon core.
Of course, the trick with DPDK isn't that it was moved out of the kernel.
It's that it's using polling-mode drivers.
A CPU sits in a literal busy loop just looking at the NIC ring buffer. Interrupts are disabled.
So that's the cost to doing that- as many cores as you have servicing the ring buffers are sitting there pegged at 100% utilization.
Doesn't really matter if it's in the
Re: (Score:2)
Fun fact, when NT came out, we switched to a 'sort of' realtime version of that, for no reason whatsoever. It nearly destroyed our business.
What do you expect? ANd NT was pretty solid back then. It is to be expected that many businesses that have made themselves dependent on Microsoft will suffer terribly over the next decades. Just look at what utterly stupid mistakes and complete failures of testing Microsoft makes these days.
Re: (Score:2)
NT pretty solid? Uh?
What versions are we talking precisely?
I remember NT servers crashing if you sneezed or looked at them funny.
Re: (Score:2)
Well, the design was. The implementation, not so much. And yes, compared to things like VMS, it was a joke.
Re: (Score:3)
> What versions are we talking precisely?
3.5.1
It was the last GUI-VMS version.
> I remember NT servers crashing if you sneezed or looked at them funny.
4.0 moved buggy drivers into kernel space "for performance".
Your wasting your breath. (Score:2)
Most here are either old enough to have an unshakable memory of MS-Windows 2.0 running on a PC-AT being slow and unstable as hell, or they've been raised on tales of how insecure, unstable and generally garbage anything written by Microsoft is.
Incidentally, I agree - 3.5.1 was NT at its best; the MVS core still showed through, and while there was plenty of room for improvement the underlying design itself was solid and robust. Like MVS, when anything went wrong you usually ended up back where you started
Re: (Score:2)
NT pretty solid? Uh? What versions are we talking precisely? I remember NT servers crashing if you sneezed or looked at them funny.
Exactly. When NT “came out”, all you needed was a fat ping packet to crash it. Hard.
L0pht and others would have never become infamous without a fuckton of NT insecurity and instability.
Realtime in mainline and that isn't the headline?? (Score:3)
AI is the hottest topic when you talk to the kernel developer #1? WTF is wrong with people?
Re: (Score:2)
AI is the hottest topic when you talk to the kernel developer #1?
They're not putting AI in the kernel.
They're using AI for code reviews and intelligent auto-complete.
Re: Realtime in mainline and that isn't the headli (Score:3, Funny)
A golden opportunity for Lennart to create systemd-LLM.
Re: Realtime in mainline and that isn't the headli (Score:4, Funny)
A golden opportunity for Lennart to create systemd-LLM.
Stop giving the guy ideas, dammit!
OTOH, maybe he could do with LLMs what he did with PulseAudio, and set the 'AI' rush back five or ten years...
Re: (Score:2)
It seems to me the sentence did not specifically relate to Linus but to what is a hot topic in development in general. The merge of RT patchset into the mainline kernel is a very important milestone for linux. But most talks at the Open Source Summit Linus attended, and presumably most development activity in the OSS community in general discussed at the conference, currently related to AI, which is what makes it the hot topic of the day.
RT Linux takes 20 years... (Score:2)
Yes, that seems slow, but it is about expected when you do solid engineering in an area that is at least partially experimental. Can also take longer. Try to do it mich faster and the results will not be good.
Re: (Score:2)
We randomly polled hwclock & RTC for 20 years. You should see the photons of Bill.
"Photons of Bill" - sounds like a cool name just waiting to be adopted by a rock band!
Re: (Score:2)
Photons of Bill sounds like an interesting astrophysical phenomena
Has a scientific paper been written on that subject yet?
Paging Sheldon Cooper !
Re: (Score:3)
ChatGPT says, "I see you're trying to write a scientific paper entitled Photons of Bill. I can help you with that!"
Rust past peak hype? (Score:1)
Linus (Score:2)
What's going to happen to Linux when Linus kicks the bucket?
Re: (Score:2)
He'll be uploaded to the Cloud and in the future when technology has advanced enough, will be downloaded into a brand-new android body.
Re:Linus (Score:4, Funny)
Systemd completely takes over. All of /etc is moved into a single binary database file and to boot securely your EFI stack has to download new keys daily.
Re: (Score:2)
So you are a;ways connected to the Internet and vulnerable to a NULL file masquerading as an EFI keyfile?
Shades of The Crowdstrike Update Disaster are now dancing through my mind.
Better than Rust? (Score:2)
Re:Better than Rust? (Score:4, Insightful)
I'm biased of course- because I've been writing software in C for 35 years... but I also adore perl, so you know I can handle ugly syntax. Rust's syntax is just bad.
Critical Deadline Missed :-( (Score:1)