Hacker News Reader

Comment by p-e-w | Hacker News Reader

There is essentially zero chance that Linux will still be in development 50 years from now, never mind 5000.

We might have strong AI by then, and surely one of its first tasks (assuming it doesn't wipe out humanity) would be to redesign our entire information technology stack from the ground up, which would make any software built by humans instantly obsolete.

But even if that somehow doesn't happen, I fully expect an operating system to emerge in the next few decades that is built around a formally verifiable safety and security model. Once it is production ready, it will dominate the market soon afterwards. The stakes associated with information technology are ever-increasing, and relying on a system written in a language where "off by one" can mean "privilege escalation" isn't sustainable in the long term.

View on HN

In 1966, IBM released its new mainframe operating system, OS/360. 56 years later, it is still actively developed, although it's been rebranded a few times – most recently to z/OS, although it is still probably best-known as MVS. I'm very confident it will still be around in 20 years, and being actively maintained – although no doubt its install base will have shrunk a lot further. By then, it will be 76.

Considering how many untold millions of devices run Linux worldwide – servers for myriad business applications, consumer electronics, motor vehicles, industrial machinery, medical devices, lab equipment, aircraft, spacecraft, satellites – if z/OS can make it to 75, why not Linux? Linux will turn 75 in 2066, and then it only need survive another 6 years to make it to 2072.

Even if some new hotness (such as Google's Fuchsia) overtakes it, its existing momentum will be enough sustain it for decades to come – and people will be willing to pay for necessary maintenance (bug fixes, support for new hardware, network protocol enhancements, etc), just like they do with countless other legacy systems today.

> I fully expect an operating system to emerge in the next few decades that is built around a formally verifiable safety and security model

We already have formally verified operating systems – the open source seL4 microkernel was proven correct in 2009. But, few have adopted it – because, whatever the theoretical advantages a formally-verified microkernel may have, they don't amount to enough in practice to justify the switching costs.

BTW z/OS has Unix subsystem, which enables it to run practically any modern Unix software, including all modern script languages, etc. which are actively used. So there's no urgent motivation to change the platform for those who're invested in it but want to also use more recent software. I think there's no reason why it couldn't live to be 100 years old - maybe not the most popular software by then, but still alive for those who need it.

I haven't personally done it, but from what I've heard, porting things to z/OS UNIX can be rather painful, due to its use of EBCDIC, and being based on an older UNIX standard (UNIX 95) which means it is missing many newer APIs (for example openat and friends).

IBM has now added to z/OS the ability to run z/Linux Docker containers (a feature they call "zCX"), which run in a hypervisor built in to the OS. It is a lot easier to just run the software under Linux, which is all ASCII/UTF-8 and has pretty up-to-date APIs (even though few Linux distributions have formal UNIX certification, Linux generally implements all the standard UNIX APIs, except for obscure things with dubious value and little use). When it is just VMs on the same machine, network communication between two operating systems can be very fast.

I don't think EBCDIC use is mandatory. It was a while ago, but I did some ports for that Unix, and I don't remember EBCDIC being an issue. Of course, z/Linux is even easier.

zSeries machines also run Linux virtual machines extremely well (e.g. the internal throughput between VMs is crazy good). Just because Linux becomes popular doesn't mean zSeries stops being relevant to the kinds of corporations it's intended for.

> We already have formally verified operating systems

But none of them are adequate replacements for today's mainstream operating systems. seL4 is a research project, not a Linux competitor. If it offered features similar to Linux, companies absolutely would switch – the cost of security vulnerabilities and their management is staggering.

This is kinda like saying if we had electric vehicles that can charge instantly and have million km range then everyone would switch. Sure, that's correct. But it totally glosses over the fact that something like that is just not possible to build in the foreseeable future. A kernel that rivals Linux in its feature set and is formally verified is even more of a bonkers proposition.

> There is essentially zero chance that Linux will still be in development 50 years from now

I doubt this. Linux is going to be around for a while just like COBOL, mainframes, and a bunch of other legacy technology.

Even today there’s not really a clear Linux competitor. It has been dominant for quite a while.

Except, forces that be won't be particularly thrilled about letting a formally verifiable, safe and secure operating system dominate the market. There's going to be serious resistance as long as humans are at the top.

Maybe strong AI can help in this regard. Or maybe it will just emulate the prejudices of its creators.

It'll have to compete with GNU Hurd 1.0.

I don't think rushing to release Hurd out by an arbitrary deadline like 2072 will be in anyone's interest.

Hundreds of thousands of people Google "how to center a div" monthly and you're talking about AI wiping out humanity or becoming a omni sentient coder? Okay....

AI has no problem centering a div

https://youtu.be/DdAhRpXE_ww

We are doomed

People google "how to center a div" because CSS is garbage and makes such basic things incredibly difficult and unintuitive. Just because the question seems simple doesn't mean it is.

Either way, I don't understand how these two issues are related.

Reply on news.ycombinator.com