The History of Windows NT 3.1 | Hacker News Reader

The History of Windows NT 3.1

abortretry.fail

78 pointsBirAdam74 comments

If you want to learn more about NT, read "Showstopper! The Breakneck Race to Create Windows NT and the Next Generation at Microsoft" by G. Pascal Zachary

Probably my favorite technology book, right up their with Soul of a New Machine.

Make that three with „The Dream Machine: J.C.R. Licklider and the Revolution That Made Computing Personal“.

I second this, I loved Showstopper, it does a great job explaining how NT is designed and why it is designed that way, and covering the personalities involved with NT.

Order placed! Thanks for the reference.

For a deep technical dive into the architecture of Windows NT 3.1, I highly recommend the book "Inside Windows NT", the very first edition, by Helen Custer. She had extensive access to the development team, including key architects, during development, and the book was published at the same time as Windows NT 3.1 was released, if I remember correctly.

One of the most enjoyable books of its kind I've ever read.

https://archive.org/details/insidewindowsnt00cust/

It's a good starting point, but deep it is not, certainly compared to other books such as (for three examples) Nagar's book on the filesystem architecture, or the Baker and Oney books on device drivers.

This article is just begging for footnotes.

> To aid in internationalization, Unicode was used top to bottom, greatly enhancing internationalization.

And we all know the story of how Unicode then went from 16-bit to 32-bit, rather spoiling things, leading to the world as it is today, with WTF-16 and code page 65001.

> NT was a fully 32 bit microkernel operating system with [...]

There was a whole lot of debate about this at the time, and the upshot is that by no contemporary measure of what it meant to be one, was Windows NT a microkernel operating system. It's sad to see this come back in potted histories 30 years after the fact, because "microkernel" was more marketing tapping into an operating systems design fad of the time rather than an accurate technical description.

> but unlike VMS [...] NT uses its own NTFS [...]

In an article calling out the similarities in other aspects, missing the marked parallels between ODS-2 and NTFS, from BITMAP.SYS versus $Bitmap to INDEXF.SYS versus $MFT, is rather silly.

> A program written for NT in 1993 can run unmodified on Windows 11 in 2023 as the entire API and userland haven’t changed while everything underneath most certainly has.

The interesting part of this sentence being in the discussion of personalities is that the Win32 personality is the only one of the original personalities for which this is true. The 16-bit OS/2 personality doesn't exist any more, the POSIX personality has been superseded four times (2 Interix systems, and then 2 WSLs), and NTVDMs went away with the move to 64-bit.

> Microsoft added the win32 API developed for NT to Windows 3.x as well [...]

Actually it was Win32s, and an article that mentions the list of caveats for the 16-bit OS/2 API on Windows NT listed in an OS2API.TXT file should really mention that Win32s came with lots of similar caveats about what parts of the Win32 API weren't supported in Win32s. Yes, just like that OS2API.TXT there was a list of what DOS+Windows 3.x did and didn't support, in the SDK. It was WIN32API.CSV. There was also a WIN32S.DAT listing all of the Win32 stuff that didn't work on DOS+Windows 3.x with Win32s.

> the Win32 personality is the only one of the original personalities for which this is true

The entire personalities thing was pretty much never used. I could never run anything OS/2 except for a couple 16-bit command line utilities, and the POSIX one was straight garbage and likely the source of many Weirdnix contest stories. It was only until Interix that you could actually source port anything worthwhile, and they quickly hit the limit that the POSIX and Win32 personalities could not speak to each other with anything other than TCP/IP. WSL1 quickly dropped the POSIX personality and switched to the Drawbridge sandbox stuff (in part to be able to access the Win32 subsystem, and to be able to skip PE binaries), and WSL2 followed by dropping the entire idea and using a then more classic HVM.

I like Win31 UI. Sure MDI was an error and hot corners would be very useful but I like how unclutered it looks and the absence of a taskbar and other always-there elements is positive. I dislike the idea of a taskbar. It takes screen real estate that would be better used by apps and when I do need the taskbar's features, those are all cramped at the bottom of the screen. Give me a hot-corner-activated win31-style program manager over a win95-style start button please.

>I dislike the idea of a taskbar. It takes screen real estate that would be better used by apps

Since the very beginning in Windows 95 the taskbar has had an "auto hide" mode if that is your concern. For the vast majority of users, having a consistent present launcher and switcher is important, as seeing a screen with no button to switch apps would be incredibly confusing.

Windows NT 3.1 and Windows NT 2000 were Microsoft's proudest achievements for OS development.

NT5 beta was more stable than anything MS released to market before, it was funny

Only if one forgets XENIX. And yes, this is pride, not necessarily success.

... I had no idea! XENIX wiki page: https://en.wikipedia.org/wiki/Xenix

It would be interesting to know more about the sources for this article because while nothing seems outright wrong, some details seem a bit off. e.g. it was called "OS/2-NT", not "NT OS/2" (unless that was the internal name at some point); while some NT features are obviously copied from VMS (i/o completion ports from $QIO), the article's mention of IRQ level is off -- that comes from the CPU design and is present in most CPUs going back a couple decades prior.

Actually, it isn't. I've implemented an IRQ level like system on Intel Architecture, and the operating system concept really doesn't match the processor's and interrupt controller's architectures very well at all until local APICs enter into the picture. Without a local APIC, one has to do tricks like re-issue interrupts with INT instructions in order to implement the NT interrupt model.

The IRQ level concept may have antecedents in earlier systems from the 1980s, but it does not originate with the Intel-based Personal Computer AT architecture of the time, with its dual 8259s, EOIs, IF, and whatnot.

Microsoft gets too little credit for NT. They were effectively 9 years ahead of Apple in developing a usable multithreaded, preemptive multitasking OS (counting Jaguar as the first usable OS X). Further, they did it in a way that (unlike on Mac) allowed software developed for the old style system to work on the new, industrial strength one.

Fifteen, if that's your metric. OS/2 1.0, which was preëmptively multitasking and multithreaded and had the Family API, came out in 1987. And XENIX was even earlier, although it wasn't multithreaded.

Hold on, let me turn my TeamOS/2 mug around so it can't see me type this...

Firstly, NeXTStep came out in 1989, and had both preemptive multitasking and threads, and OS X 1.0 didn't really improve on NeXTStep in that area at all.

Secondly, while OS/2 1.0 had threads and preemption, it was text-only and had to run on IBM AT 80286 PCs so most of that technological might was wasted. It would be 1990 before it had a real GUI. And it would be nearly 1992 before OS/2 was 32-bit.

It's eduction's choice of metric, not mine, and by that "multithreaded, preemptive multitasking OS" metric, OS/2 1.0 qualifies. eduction didn't say anything about graphics.

Presentation Manager was available in 1988, by the way, not 1990. Petzold's book on it came out in 1989. First (and only) edition sitting on my bookshelf as I type this.

Amiga OS had preemptive multitasking in 1985, in 256k of RAM. No memory protection though. I’d consider Amiga tasks light enough to be threads, since they shared memory.

They also flat out stole MICA source code and questionably IP. If DEC hadn't been so weak and wanted to boorlick Microsoft they could have made a bigger deal of that.

Allegedly. Though, DEC had pissed away $Billions on its VAX 9000 ECL mainframe, and why fight a costly legal battle?

[0] https://www.itprotoday.com/compute-engines/death-alpha-nt

[1] https://techmonitor.ai/technology/dec_forced_microsoft_into_...

For sure, they made a lot of really bad choices. VAX9000 is maybe the biggest one for sure.

Using all that money to push PRISM/ALPHA and MICA into workstation and server and large server space would have yields them better results. That said, I'm not sure if DEC could have survived with that strategy either.

What it would have done is make it so that Windows NT ends up not existing or existing in a very, very, very different way, likely in a far worse way. Lets remember that Gates didn't even believe in Virtual Memory.

NT wasn't ready for general consumers until XP in 2001. It took Apple a similarly long time to whip NextStep OS into shape.

I always considered Windows 2000 to have that honor. Had the licensing been there, it could have gone mainstream. Much better alternative than Windows ME.

I remember very much how XP was universally panned versus Win2k when it arrived, but XP just boots significantly faster even on the same hardware. I think it's the first one who really had any chance to replace 9x/ME.

When was NT able to play Windows 95/DOS games? Was it 2000 or XP? I remember 2000 being able to do it.

I remember playing Half Life on 2K, and it had Direct-X support.

I don't think 2K could do DOS games, as they usually talked to the hardware.

They’d work through NTVDMs.

Shortly after the release of ‘95. Microsoft’s DirectX was made available for both NT and CE. NT was able to run anything for DOS or Windows pretty much right away, though some bugs were evident in games.

I agree with you. I actually tried to run Win 2000 as an OS on my personal machine for a while (from a license I 'borrowed' from my work).

The stability was amazing, but there were still some compatibility issues with games and so on.

This is rewriting history with present hindsight. NT was a hugely bloated OS (imagine an OS today that requires literally 4x the RAM of your mainstream OS just to boot) and almost killed Microsoft several times over until computing resources caught up. And still, a shitton of sacrifices had to be made to the original design until anyone outside of a server farm could run it.

> imagine an OS today that requires literally 4x the RAM of your mainstream OS

This is just every OS when a new version comes out... sad!

WinXP needed 64MB RAM and was considered hugely bloated at the time.

No, it's not even comparable. Windows NT came in the early 90s and required a whopping 16MB of RAM (at a time where most home PCs could not even address that much). Windows 9x came _years later_ and only required 4MB. And 9x was already considered bloated (new GUI et al).

To be pedantic, Windows 95 needed 8MB to be usable without constantly swapping.

Well, to be pedantic, it's not like NT would be usable with 16MB either; I remember the desktop taking 5 minutes to show and the HDD constantly thrashing.

EDIT: Actually, I just checked, and NT 3.5 opens in around 8 minutes and uses 20MiB of swap, when booted with 12MB.

Interesting. What is it about NT that causes such a high memory requirement?

Well, it wasn't meant for home PCs....

It was marketed and sold as a business OS. I agree with the Op (OPs?) that it was revolutionary.

Just before Apple bought NeXT, Windows NT was a fully functional, usable workstation OS and clearly the established and future foundation of Microsoft's OS efforts. Apple had bupkis. NT vs bupkis looks really good for NT, hindsight or not! Apple pulled off a miraculous recovery but if anything, time and Apple's subsequent success has made the stark contrast then hazier.

Both of your posts are right. Windows NT was way ahead of the competition, and yes the new design was "bloated" simply because it wasn't hand-optimized for any given architecture the way Windows classic and classic Mac OS were (if that's bloated, I have bad news about Linux!) And you are right that it's unfair to compare it to Mac, because it was designed for "server farms" (or what passed as server farms back then) and it wasn't until XP that we saw it make the jump to a consumer product.

> If that's bloated, I have bad news about Linux!

This remark points you don't realize how bloated NT truly was.

In the 2000s I could still run Linux _and XFree86_ in under 4MB of RAM.

NT in 1993 would just BSOD with less than 10MB of RAM.

Which clearly didn’t matter to their target market, which you didn’t represent.

The GP talks about NT, Mac, etc. in the context of home computers, a market which (by virtue of being ridiculously bloated) NT couldn't actually target for an entire decade (such market was then using OSes where the minimal requirement was more like 0.5MB of RAM). At the very least it is unfair to claim it was ahead of the other OSes in that market, since it literally couldn't even think of competing with them.

The parent poster claimed the bloat was because of portability and other metrics, but this is also not true since other very much portable OSes from the next decade still had lesser minimum requirements.

NT was not such a good design as it is often claimed. It was a clone of a esoteric big iron OS that they were trying to retrofit on much smaller devices. They had to backpedal on everything from microkernel to personalities in order to be able to actually market it to "their target market".

Yet despite all these shortcomings it, and its descendants, were, and continue to be, wildly commercially successful.

I genuinely don’t understand all this looking down on what, in hindsight, clearly worked.

NT was always marketed and priced for the professional & enterprise markets, so the whole “but what about home computers at the time” rings hollow.

NT ushered in a new era of business computing and networking, and more importantly a new era of thinking about and working with technology in the workplace, something its competitors at the time couldn’t manage. Even Novell netware, dearly missed, got steamrolled by NT. You can pooh-pooh NT all you like, and talk about what a shitty clone it was, and how poor the technology was.

NT clearly won, at the time.

Am I missing something? Doubling the memory requirement for a piece of software by adding layers of abstraction isn't uncommon, no?

It's not just doubling, but literally going out of reach for most people in the home market. A comparison with other home OSes of the era is therefore unfair.

I think I object to the term "bloat" as used there. To me, "bloat" means that something is large for no reason. If OS/2 and NT were able to leapfrog the home PC market by using a high-level language, that seems like an acceptable engineering tradeoff.

I was running Linux pretty early on (1993, SLS Linux, Kernel 0.99.x days.) X barely loaded in 4 megs of RAM back then, so no way was it running in 4 megs in the early 2000's. You are exaggerating. By the the 2000's, desktops with 256+ megs of RAM were common. The bloat across all OSes had begun.

> so no way was it running in 4 megs in the early 2000's. You are exaggerating.

No, I am not. I used to do that with at least 2.2 kernels. https://tldp.org/HOWTO/4mb-Laptops.html

But besides, I know 4MB in 2000s was low enough to be awkward. The point is that _it could_. NT in 93 could not even boot with 8MB RAM, and then it was definitely not low.

From https://tldp.org/HOWTO/4mb-Laptops-1.html#ss1.5 : “It isn't going to run X”

This seems to confirm my statement.

I still have the machine. Removing all but 4M and booting XFree VGA with an xterm doesn't even use more than 500KiB of swap, so it's actually usable. So is 95.

On the same machine, NT 3.5 BSODs. With 12MB is the first I can get it to boot, albeit it complains that some services failed to start, and generally thrashes like hell, using 20MB of swap...

You can even try muLinux yourself, albeit it's kernel 2.0, not 2.2, and it opens some X11 utils that bump swap usage to 2MB (not very usable).

Windows NT was practically the perfectly timed investment because Microsoft also had the MS-DOS + Windows product to tide over the mass market while Moore’s law ensured it would eventually catch up with the much better foundation in NT. Bill Gates deserves credit for having the foresight to hire in 1988 the best people to design an operating system that wouldn’t shine until the hardware of 1998.

There’s no way NT “almost killed Microsoft” because Windows 3.x was a money printer. And if that hadn’t happened, Microsoft wouldn’t have exited the OS/2 relationship.

Windows NT was originally meant to be OS/2 3.0. In the worst-case scenario they could have stuck with that plan and played the role IBM had designed for them, which would also have been lucrative (although nothing like the bonanza of the desktop monopoly they had by the late 1990s).

> They were effectively 9 years ahead of Apple in developing a usable multithreaded, preemptive multitasking OS…

But 23 years behind OS-9, which I happily ran on my CoCo 9 years before NT was born.

https://en.wikipedia.org/wiki/OS-9

https://colorcomputerarchive.com/repo/Documents/Books/OS-9%2...

Mad props for OS9 on an 8-bit microprocessor.

The Motorola 6809 was a nice architecture for UNIX-like systems... if we accept the 16-bit memory address space.

Tangential, but I'm starting to genuinely hate Substack now. This time I've just closed the page instead of dismissing the newsletter modal for the 10000th time. Why are authors still using it?

Add the "annoyances" list in your adblocker. (What newsletter modal?)

https://imgur.com/a/MyeHnnY

And I dunno about GP but I would prefer what the websites would be usable without an adblocker, hazmat suit and poison testers.

I have always considered substack to be the reddit of blogs. That is not a compliment.

Well, it is still a better Medium.

Why was the title changed?

Had Windows NT fully embraced the POSIX subsystem, maybe I wouldn't have bothered to install Slackware 2.0.

Microsoft most likely repents of selling away Xenix, and not fully embracing POSIX at the right time.

I remember talking to some MS dude (from the NT group) at a conference, I think in 1993, where I said something to the effect I was waiting for them to add sockets support to the posix subsystem. He response was almost to laugh in my face. Nice to be proven correct 30 years later.

Yes, large scale adoption of OS X has proven that the large majority cares more about POSIX/UNIX in general, than GNU/Linux in particular.

WSL ended up being an easier way, than fixing SUA.

I thought they added Posix support because certain government contracts required it.

Indeed, hence only the minimum as required by DoD.

Eventually there was an evolution with SUA, but never with a big commitment level.

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

"Every program attempts to expand until it can read mail. Those programs which cannot so expand are replaced by ones which can." -JWZ

It seems like Every* operating system adds some sort of Posix compatibility..

Some of IBM Mainframe OS's, AS/400 via PASE; HP's MPE->MPE/IX, VMS->OpenVMS, Stratus VOS, Tandem(HPE) Guardian OSS, and any number of clones/embedded systems, etc.

*I don't mean that literally.

Mostly because C and UNIX symbiosis, and ANSI C being too little.

So POSIX ends up being the rest that everyone reaches out for when using C.

The irony being that still isn't enough to get rid of #ifdef spaghetti, due to the things POSIX leaves out to each implementation.

I've installed Windows 3.1 a few times for fun. I like it better than current Windows. However, I navigated away from the article due to the newsletter popup.

Windows 3.1 and Windows NT 3.1 are about as similar as Javascript and Java though ;)

(I still have a soft spot for WinNT 4.0, this was the first Windows that wasn't a complete joke when coming from an Amiga)

They are identical UI-wise, but not very similar under the hood.

As someone that spent lots of time with Amigas and Demoscene, the complete joke part is a bit overboard, considering the multitasking limitations and the mix of Assembly, BCPL and C.

Reply on news.ycombinator.com