4 ms·

> 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 traffi

> 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.

zaarn 4y ago
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"
- sigmoid10 4y ago
  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.
  - zaarn 4y ago
    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.
  - wongarsu 4y ago
    See also this whole radio-altimeters vs 5G debacle that wouldn't have happened if radio altimeters used a reasonably modern design.
coldtea 4y ago
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.
Steltek 4y ago
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.
windexh8er 4y ago
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.
hedora 4y ago
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.
prirun 4y ago
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.
RustyRussell 4y ago
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!

zaarn 4y ago

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"

sigmoid10 4y ago
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.
- zaarn 4y ago
  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.
- wongarsu 4y ago
  See also this whole radio-altimeters vs 5G debacle that wouldn't have happened if radio altimeters used a reasonably modern design.

coldtea 4y ago

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.

Steltek 4y ago

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.

windexh8er 4y ago

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.

hedora 4y ago

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.

prirun 4y ago

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.

RustyRussell 4y ago

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!