Hacker News new | ask | show | jobs
by evanjrowley 379 days ago
If it works, then don't change it.

I would trust a floppy-powered Windows 95 system over the horror show that passes for common operating systems in 2025.

What will they think of next? Adding AI to the ATC system?

11 comments

Operating systems have gotten a whole lot more reliable since Windows 95. The way I remember it, Windows 98 would regularly corrupt itself and need to be manually reinstalled. I'd done it so many times that I could pretty much recite the license key from memory. Modern Linux is rock solid. Even Windows 10 is very stable. They might be 'bloated', but modern OS's are way, way more stable.
corrupt itself and need to be manually reinstalled

In my experience that's normally the fault of third-party software, and otherwise quite easy to determine and avoid/fix. Now OSes with more protections just hide those bugs, causing most software to regress to a barely-working state.

I ran 98SE as a daily driver from late 1999 until 2010, and it was reinstalled at most 3 times, not even coinciding with hardware upgrades.

Or of just a power outage or driver causing a loss of write back cache.

95 and 98 and ME crashed on a regular basis. I specifically remember upgrading from ME to XP and being so happy with the massively improved stability of the NT kernel over the 9x kernels.

If you think that's 9x was stable and reliable, you may be thinking very nostalgicly.

I am not so sure. I've ran 98 on bad hardware, and it crashed regularly. So much so, that I installed linux on it already in 1998, and that was much more stable. It only crashed now and then. No doubt in both cases the poor hardware was the cause of it.

Anyway, two years later I got a brand-new laptop with good hardware that was running 98se. As far as I remember, it didn't crash during normal usage. By then I was studying computer science, and would sometimes write or run programs that would make it crash, but that was on me. I did dual boot in Linux, and that didn't have any problems on that machine either.

Fun fact, I still have that laptop, it's over 25 years old now, but it still works and runs Windows 98se!

Or a modem driver reading the stream and writing shit - I still have some of those burping mp3s. But if you blame this solely on the OS then you may be thinking very nostalgically too.

Hell, it most of the time worked on some combo of the cheapest parts - modern systems wont even get to UEFI boot part on the parts of the same quality.

    > If you think that's 9x was stable and reliable, you may be thinking very nostalgicly.
I agree. Remember Plug'n'Play? It was so bad that we used to call it Plug'n'Pray. It frequently caused PC crashes. Modern OSes are a miracle in how stable they are with drivers.
Win9x was stable and reliable. It was the drivers that were not. WHQL wasn't invented yet.

I'm sure ATC systems were properly tested, including the drivers. Don't compare that with cheap consumer PCs that we had.

Windows 95 allowed programs to disable system interrupts. No protection on cli instruction. Careful with those rose-colored glasses.
So what?

ATC systems are definitely not going to have users installing random software at any time.

It's stable if you have a controlled environment.

I've seen enough stories of power outages permanently damaging SSDs, that if you have bad power from your utilities provider and can't get them to fix it, then I recommend investing in a UPS.

Operating systems were always more reliable than Windows95 from the day it was introduced. Protected memory and process privilege were not exactly unknown when DEC was selling VMS. Or for that matter when Microsoft was selling Windows NT. That the FAA cheaped out then, choosing an inferior system with no technical merit, is prelude to the current problem.
I've noticed that operating systems can get very flaky when the disk space gets tight. It seems that too much code does not check for disk full write failures.
It was still very much like modern systems. If you didn't install, uninstall, or aggressively reconfigure things they were pretty stable, and controlled changes could be achieved. Some of the problem though was that the systems required a lot of that to do anything fun with them at home.
> If it works, then don't change it.

As the article points out: the hardware is at risk of physically failing and it’s getting harder to replace like for like. That’s the reason for looking at an upgrade. Hell, even turning the machines off to replace them is a challenge since some systems need to run 24/7!

If the machine is that old, couldn't an emulator be used to migrate to modern hardware?
Not necessarily. For example, if there is custom hardware used for communications with other systems, such as radar for example, there might be specific timing and latency requirements that could be difficult to meet under emulation.

The most recent Dolphin Emulator post referenced a bug they had where memory cards were written to too quickly under the emulator (and even on actual hardware if you had memory cards that were sufficiently fast) which caused some games problems because they did not expect save files to be written so quickly. Imagine things like that, but where the worst case isn't having Wind Waker hang while saving, but planes crashing.

A long time ago, ..., so add a huge IIRC for the details.

Anyway, a long time ago, in a physic lab class, we had a custom spectrometer device that was controlled using the printer port. (Probably a light, a detector, a diffraction grid and a steppers motor, and perhaps some weird card as an interface to the computer.) It was controlled by a custom Pascal program in the DOS command line.

It was connected to a Win 95 computer that died and we had to replace it with a Win 2000 one. The problem is that anyone can read/write the printer port in Win 95 but that's not possible in Win 2000. [1] We had to make some magic, probably rewrite the program and use some weird device driver to write to the printer port (IIRC again). And multitasking messed the timing. At the end it worked, but upgrading the computer was not easy.

[1] If that makes no sense, replace that with a Win 3.1 to Win 95 transition, but I think it was not that old.

I had similar many years ago in a custom paint shop. We had an expensive colorimeter that interfaced over serial with a program that ran in DOS.

When the pysical computer gave out, I replaced it with a reasonably new one but instead of using a modern OS I installed MS-DOS in order to get it up and running as reliably and quickly as possible.

If I were doing the same today, I'd likely get a new computer and install FreeDOS.

Back when I was designing electronic circuits, the rule was to design for minimum speed, but faster speed should not cause a failure. The rationale was that newer parts were usually faster, and the older parts disappeared.

Of course, nothing can prevent poorly designed code and hardware.

The speed thing was just an example that easily came to mind. I can imagine there are other kinds of analog vs digital interactions that might be occurring that may not be easily replicated under emulation. Especially with a system that grew somewhat organically over the last half-century.

Emulation is likely possible, probably for many of the systems involved, but this is not a field where bugs, especially ones introduced due to emulation, would be easily acceptable.

> Of course, nothing can prevent poorly designed code and hardware.

Agreed, but the reality is that here, trying to fix things and ending up breaking them can and probably will kill people.

These kinds of risks are always present with anything involved with airplanes.

Being terrified of progress means you're likely to kill far more people.

I would not be surprised if many of these systems rely on several ISA cards with proprietary protocols and drivers.

I am only aware of a single modern-ish motherboard with ISA, the MS-98A9, and it only supports Intel 3rd Gen Core series CPUs.

That said, if it was a large enough project, reverse engineering and re-implementing using modern components would likely be feasible. Turning each of these into network services handled by something more akin to a RaspberryPi could modernize the data sources while providing a sustainable and modular replacement strategy. The problem is that its not "sexy" enough to get a major government project, and it would not grease the correct palms that a multi-billion dollar next-gen complex proprietary replacement would.

I'm sorry, but this just sounds like quitting before you start.

For example, I've looked into emulating DOS, because I don't like the existing emulators. But I don't need to emulate floppy disk drives, or their drivers, or their hardware cards. I just write some software that can fake a disk drive and hook it to the I/O interrupts. That would be a simple project.

There's nothing sophisticated about DOS.

In hindsight, I'm baffled that it took many many years for people to develop clones of DOS.

For example, EDLIN. A trivial program. I'd write it in a high level language like D, get it to work, then hand-translate it to asm. The executable loader is absurdly simple. And so on.

That conjures up memories of Joel Spolsky's cautionary tale about lava flow from his talk at WeAreDevelopers 2019: https://youtu.be/tWKh95Kio38?feature=shared&t=311
If it uses any analog hardware it makes emulation much harder.
If it uses any digital hardware directly accessed by software (something that Win9x still allowed, see DOS games w/ ISA sound cards), it makes virtualization impossible.
Why is that? Doesn't virtualization virtualize the hardware? I'm not sure why virtualization is impossible there, as QEMU can run plenty of old DOS games.
Yes, but what use is a virtual radar, or a virtual radio?
No, pretending to behave like other digital hardware is exactly what virtualization does.
My biggest issue is probably the reliability of floppies over time, and the lack of manufacturers left out there, with how few orgs still use them.
Yes, we know that floppy disks and drives will wear out, and they have few if any sources for new repair parts. So the fact that the system is still more or less working today doesn't mean it isn't doomed and needs to be replaced before experiencing a catastrophic unrecoverable failure.
Sure, but the timeline before it’s a significant issue easily extends another 20 years.
Maybe. The problem is that you don't want to have to deal with that maybe in some applications.
40 years would be a maybe, 20 can be guaranteed by buying existing stock.

Even 8 inch floppies + drives are still available and those got abandoned a decade before 5.25” let alone 3.5” disks.

Why hasn’t someone come along to manufacture and sell floppy’s to these industries for $15 each? Or even $50?

Seems the market has demand, at least in the mid-term.

1. There are still some (but few) new old-stock floppies in existence. https://www.tomshardware.com/news/floppy-disk-still-has-life

2. There are floppy emulators that replicate the functionality of floppy drives with flash

3. The above two probably absorb all of the demand today, but even if they didn't, the volume is so low that fixed manufacturing costs per unit could likely push unit prices well beyond even $50. The tooling for factories often costs millions and unless you are selling in high volume, you will have quite a high fixed cost per unit.

Because it wouldn't be profitable? How many do you think they could sell to a dying market, and what would those manufacturing costs be? What experts could you tap who know this space? they are all gone
The article says it will take (at least) 4 years just to roll it out. 20 years isn't much... legroom.
The point is it’s enough legroom to be reasonably cautious in the rollout rather than needing to get a big contractor to do a major and therefore expensive push.
I read some years ago - IIRC the letters pages of BYTE, which dates it - about a critical factory control system in a company somewhere running on an IBM XT. The MFM drive had started to show some errors, so they got in touch with IBM, who being IBM, did not have any drives in stock (they'd stopped making them 15 years previously), but could retool a manufacturing line and make some. They offered to do it for $250k/drive. The company paid up.

That was cheaper at that time, than modernising that system. But it's clearly not long-term scalable.

I've heard of S/360s in KTLO mode in basements keeping banks running. Teams of people slowly crafting COBOL to get new features in at a cost of thousand of dollars a day each, and it "still works". But from a risk point of view, this is also ridiculous.

Safety critical systems have different economics. Yes, you can keep the floppy systems going, but the cost of keeping them going is rising exponentially each year, and at some point a failure will cost one or more airliners full of civilians and the blame will be put on not having a reasonable upgrade policy.

Sometimes you have to fix things before they stop working, or the cost is not just eyewateringly expensive in terms of dollars, but of human lives too.

IBM mainframes can run software written in the 1960s without modification. There’s no reason anyone would keep using an obsolete mainframe, and IBM usually leased them anyway and would refuse to support obsoleted machines.
You clearly don’t know what has been happening in the World of S/360 (and similar), support contracts in recent years.

Costs are rising heavily. IBM sold off most of that business, to people who don’t really want it as the skill base to support it is retiring and it’s too expensive to easily replace. This has been going on for a couple of decades, but it’s now gaining more and more pace.

Let's be a little more reasonable. I don't think anyone is saying we need AI. There are numerous other technological advances between floppy drives and AI that our air traffic control system could benefit.
If you have been reading the news lately you’ll know that the ATC system absolutely does not “work”.
What, you don't want a Next.js React and Node.js app with 3000 dependencies commanding your flights?
Isn't it what they use on the Space X Dragon controls?
The thought of npm in space is making we sweaty. Let's hope it's not near anything that's life safety critical.
Does it work? Sure. You have to ask more questions. How much does it cost to keep it working? How much would it cost to upgrade? If we do nothing, along what sort of timeline can we expect it to stop working, or become cost prohibitive to maintain?
Well also, 20 years is less time then you think. For a system of this magnitude, deploying the replacement could easily take 5 years to get all the way through to full completion. So that's 1/4 of your runway gone right there.

Every year you delay is pushing that lower, and then there's whether the funding is available because you're in fairweather economic conditions or if crisis will happen concordantly with some other crisis (I.e. do you want to be stuck replacing air traffic control systems in a rush because some war has wiped out the floppy supply chain right as your air logistics is a critical issue?)

They are changing it for publicity reasons. It has nothing to do with whether or not it works.

This whole thing is being done as a reaction to this video:

https://youtu.be/YeABJbvcJ_k?t=1540

The article completed skipped over this. This video was released literally a week ago and is completely mocking the FAA. Floppy disks are a big joke in this video.

I'm not even sure if it's irony or sincere backward thinking of the same type that locked us with leaded fuel in post-war design lycomings.
Just vibe code a new ATC system.