Hacker News Reader

Comment by uluyol | Hacker News Reader

> Through Mesa and Gallium3D, we benefit from thirty years of OpenGL driver development, with common code translating OpenGL into the much simpler Gallium3D. Thanks to the incredible engineering of NIR, Mesa, and Gallium3D, our ragtag team of reverse-engineers can focus on what’s left: the Apple hardware.

The Linux graphics folks have really achieved something.

View on HN

In the case of AMD, the Gallium-based open-source OpenGL drivers also manage to compete very well with AMD's proprietary OpenGL drivers, both in terms of performance and features, despite the fact that the proprietary drivers are much more mature. See: https://www.phoronix.com/review/radeon-spvp2020-linux

There's even a working Direct3D 9 driver based on Gallium3D. With an appropriately patched WINE, you can use it to run old Windows games quite nicely even on integrated graphics or with low core count CPUs. Here's some reporting: https://www.phoronix.com/news/Gallium-Nine-Better-2021

I hope that with NVIDIA's new open-source kernel drivers, now Nouveau can push forward and get good performance like Mesa's open-source drivers for AMD. That'd be awesome, and it could pave the way for mainlining NVIDIA's new kernel driver.

I remember reading about David Miller, I think, porting Linux to SPARC.

That was an amazing feat, this is also very impressive.

I think at some point Linux was as good as Solaris, if not better, on SPARC systems.

I really wish this were true at some point for Apple hardware.

> I think at some point Linux was as good as Solaris, if not better, on SPARC systems.

That's a bit of a stretch. In fact, the last time I was in an air traffic control tower, it was Solaris, not Linux, in active use by the controllers themselves.

I don't think someone being stuck with Solaris for unknown reasons is sufficient to declare that Solaris is better or equal to Linux. For the ATC, it could simply be a matter of "The application we run only support Solaris"

It's aviation. Even if there was a superior solution and it was readily available, it would take decades to get through certifications and before it became widely accepted or even the new default. See GPS vs radio navigation or just analog radio communications.

Yeah that would be on top of it all. For the ATC to use Linux on something, there would have to be a competing application that runs on Linux, which has to be certified and then migrated towards. That could take a decade or two.

See also this whole radio-altimeters vs 5G debacle that wouldn't have happened if radio altimeters used a reasonably modern design.

That would 100% be based on how those systems were acquired and where they were tested and delivered originally (e.g. through a tender and some Sun or Oracle support deal), not whether Linux is better than Solaris on that hardware on some particular metric, and even less so if 2022 Linux is better than Solaris on that hardware.

It would also have to go through many series of bureucracy, compliance requirements, and such to be installed in the first place. Even a point update to the next Solaris version could be a year long process, complete with several staging systems and so on.

Last but not least, it would 100% be tied to the ATC software run there, and under what OS it was developed and tested (which, if it was pre-2000, would more likely be some commercial UNIX like Solaris, considered - and being - more mature and supported for such use then).

An airport management wont just go and reinstall some mission critical software they got through a specific contract deal. And ATC software wont be just some GitHub repo you recompile and build for Linux.

But that's almost totally unrelated to how well Solaris vs Linux runs on the machine.

Anything touched by the FAA wasn't considered on technical merit, it was pure inertia and being terrified of any change, no matter how warranted. C.f. leaded gas for general aviation.

You're probably right. I worked for Lockheed in a facility that was ATC and MS2. I was on the MS2 side, but our program was ATC for NATO countries (war and peacetime operations).

All of the ATC applications were built on Solaris running SPARC. Most of the developers were familiar with Linux at the time, this was back in the early 2000s. But even then many of the devs wanted to migrate the platform to Linux for a number of reasons. The cost and time, however, for acceptance testing on Linux would have eaten the budget alive.

So instead pieces and parts that were readily accessible in Linux were ported to SPARC. I remember getting a new requirement for GPS time (previously the system had only used Rubidium oscillators) and working with one of the devs on getting OTS hardware working with the ported code. The further along the program went and newer features were added to the scope the more this happened. But everything in the UI was based on CDE and some SPARC specific libraries for the UI. The HMI was written in ADA.

Since these systems have so many requirements just swapping out the OS would be a major overhaul and I'm not actually sure that Linux would even be the right choice.

Ignoring the air traffic thing, I agree it's a stretch.

Back when Sun was still in business, Linux's SMP support was still in its infancy, futexes were not a thing, and the pthreads documentation was nonexistent.

On top of that, fsync was actually broken on ext2/3. Also, there was a single kernel level lock per file, so you couldn't have two CPUs seeking in the same file at the same time (oracle recommended using a block device instead of a filesystem, and provided a list of changes to the kernel source you needed to make manually if you wanted Oracle on Linux to be supported.)

None of this mattered of course, but the Linux kernel certainly wasn't "better" than Solaris back then.

Even if it was running on Linux, it would probably be some ancient version of Linux that past gobs of certifications at some point. There likely wouldn't be any OS upgrade until another lengthy certification was done, ie, never.

At USENIX in 1997 I went to a talk by a young David S Miller and Miguel de Icaza on porting Linux to the Ultrasparc. That talk (regretfully unrecorded) more than anything else convinced me that I had to work on Linux.

Throughout the talk they showed Linux vs Solaris performance, and talked about their optimizations. It was lmbench (Larry McVoy the author was ex-Sun, which adds something) that, and by the end they beat Solaris on every result.

Obviously this was impressive, but it's hard for me to clearly express how much this shook my assumptions about how to build good software. A bunch of students shouldn't have been able to beat Sun on their own hardware in anything!

You're not telling the most important part of that story: https://archive.is/KgAYd

https://en.wikipedia.org/wiki/David_S._Miller

https://en.wikipedia.org/wiki/Bryan_Cantrill

I didn't want to be a jerk to Dave.

These days I think even Cantrill regrets being a jerk back then.

The weird thing to me is that I don't understand how anyone is expected to believe that Bryan Cantrill is somehow less of an obsessive nerd than Dave Miller or anyone else. Like what character was he playing when he wrote that line??

Maybe he’s admitting defeat with an ironic insult. I say, “shut up, nerd,” to coworkers when I’m wrong in a technical debate. We’re all nerds to some extent, so it’s funny.

On the other hand, maybe he was playing the guy who lashes out when his feelings get hurt after his work was publicly eviscerated.

For whatever it's worth, it was a Saturday Night Live reference -- and I absolutely regret it.[0]

[0] https://www.reddit.com/r/IAmA/comments/31ny87/comment/cq3e4y...

If only you'd also included one earlier line from the same skit¹:

> You've turned an enjoyable little job that I did— for a few years— into a colossal waste of time!

it would have balanced out the later one!

Thanks for making it make sense :)

1: https://trekmovie.com/2021/09/06/william-shatner-defends-198...

Reply on news.ycombinator.com