> 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.
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.
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.
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.
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.
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!