The vagaries of 1990s 32-bit Windows networking
liam-on-linux.livejournal.com
liam-on-linux.livejournal.com
Aargh. I don't think even the more advanced Linux distros have reached that level of complexity yet. Although I suppose Prolog is a natural solution for constraint solving .. why do you need a constraint solver anyway?
(Linked deep dive https://web.archive.org/web/20040603192757/research.microsof... )
No, but they do have parallel config utilities fighting each other in the case of Linux mint. Took me a few hours to figure out why I was unable to switch from a static IP to DHCP.
Windows 10 had this as as well: the new user friendly ui can show dhcp while the interface actually is statically assigned An ip address. Took me while. Seems they dropped not only any consistency but also the prolog part in the new ui.
There was more prolog hidden underneath network operating systems than people think, too.
WinNT network config covers a lot more than people are aware even if you keep just to TCP/IP - how many people are aware that all WinNT >=5.0 systems are OSFPv2 routers? There's also of course optional RIP listener, or the part where WinNT >=6.0 is IPv6-first.
When the claim was made, you could end up working with few major network protocol families, and while vendors failure was leading to ending GOSIP it was still unclear if TCP/IP would be winner (I think TUBA - aka IPv9 - was still in the running when the prolog part of network configuration was done), you had huge installed base of IPX/SPX (that didn't really go away till early 2000s), you had NetBEUI support for dealing with LanManager variants, there was Banyan VINES you might have to cooperate with (especially if you tried to sell to USMC, iirc), there was X.25, FrameRelay and other WAN specific protocols you might need to support, there was AppleTalk, there was ATM (not just IP over ATM), there was support for dealing with IBM Mainframe networks...
All of the above also had presence in the Unix market, and given 1993 release date, all of that was present in some ways - and Microsoft could capitalize on "we can easily support all of that".
Most unices of the time barely supported anything else other than tcp plus whatever vendor specific stack they might need.
Windows networking stack was easily extensible, something that to this day Berkeley-derived ones have issues with (Sockets, being "temporary hack", lacked features to handle multi-stack well). A non-trivial reason I heard for popularity of Solaris in certain telcos was that it offered XTI interface, meaning much easier access to ISO/OSI related protocols present in many areas of telco gear.
I even checked online to ensure I didn't typo-it, except I ended up (apparently) on page that had the typo...
We found that with Navigator network performance through the proxy was completely terrible--unless you were waving the mouse around. If you were waving the mouse around then it was fine.
I don't remember all the details, but I think it turned out that this only was the case when Navigator was going through a proxy on 127.0.0.1. If we changed it to use the host's external IP address then it worked fine regardless of what you were doing with the mouse.
Internet Explorer did not have this issue. We had another thing we were working on that involved an email proxy, and no email clients we tried had any oddities with 127.0.0.1 proxies.
Later, we switched away from explicitly configuring browsers to use the proxy. Instead we switched to an LSP [1]. 127.0.0.1 then worked fine on Navigator, so whatever the issue was it only happened when Navigator knew it was talking to 127.0.0.1.
"Fair" scheduling in these situations is a nightmare that I think I've only seen BeOS and the research OS Nemesis handle correctly.
(It can be enabled via a compile-time define because my OCD kept me from shipping an incomplete port.)
Can anyone confirm / debunk this?
I've noticed that younger crowds seem to treat Windows 98 as a big revolutionary release, attributing many features that were actually introduced in Windows 95 as being uniquely new to 98; rather than Windows 98 being a rather minor incremental upgrade to 95 as it actually was.
Not really sure why this phenomenon has happened. Maybe people had glimpses of Windows 98 in childhood and don't know what the prior version was like? or that Windows 98 probably had a longer lifespan than 95 did?
From the user perspective, almost the only differences are Active Desktop and titlebar gradients. FAT32 and USB are other tangible differences, though it's worth noting that 95B also had those features, it just wasn't available at retail.
Of course people would build their own computers from scratch in those days too and pirating Windows 95B (later, 95C) would be fairly common. It had big benefits while any Windows 95 box at the store was only 95A.
But even including those features, coming so soon off the heels of the 3.11->95 change, it certainly felt a minor change to me.
That said, 98 was the last of the Dos era (the less said about ME the better), so it did persist until well into the 00s for people playing games.
And of course you've got the other "grown up" line of NT 3.51>4.0>2000
From this user's perspective, Win98SE was far more stable and configurable, had vastly superior driver support and ran more types of software better. It was fantastic for everything from internet to audio editing to gaming at a time when I was doing a lot of tinkering. I struggle to think of specifics now, but I know that when I was using it a lot I ran across numerous things I valued that I couldn't do on Win95.
> Windows 98 probably had a longer lifespan than 95 did?
It definitely did in my experience and that of several tech-inclined folks I was adjacent to. In households with kids who paid attention to details like which OS was running, I suspect greater numbers of those kids were exposed to 98 than 95.
> From this user's perspective, Win98SE was far more stable and configurable,
> had vastly superior driver support and ran more types of software better.
I worked with it, installed it and supported it at the time, and it really really wasn't.
Serious business users ran NT 4. In the trade we nicknamed W98 as GameOS.
I didn't put in a lot of Win95, because it was before I went freelance. Then my work box ran NT 3.51, but a lot of the office ran W95 and I had to support it. So I did get quite a lot of experience on it.
Later on, though, I put in W98 & W98SE for people with cheap & nasty PCs that couldn't run NT, and I supported a lot of boxes that were out there.
It did have somewhat better driver support, yes, but every version of Windows ever has had better driver support than the preceding version.
It was not significantly more stable, and on low-end or heavily-loaded boxes, I'd say _less_ stable.
More types of software? About the same, I'd say. Better? On a narrow range of hardware, maybe -- things were changing fast then and by '99 or so new kit couldn't run 95 any more.
It was a gamers' OS.
The much-maligned WinME had better hardware support still -- for example, proper working Firewire support -- but by then everyone hated it. After a few updates it was a very stable release, it looked nice, it was a bit cleaner and faster to boot and to shut down, and it supported more memory.
There was a particular period when very cheap, or old, PCs couldn't usefully run Win2K let alone XP, which really wanted 256MB of RAM or more, a big fast DMA EIDE or UltraIDE hard disk and so on, when WinME could deliver a decent internet experience on a single-core box with between about 80 to 128 MB of RAM on the low end to say 192 MB on the high end -- kit which gave a rotten experience with 2K or XP.
But 98 wasn't a particularly sweet spot in my recollection.
In 1993 I worked in the City of London, where NT 3.1 was used in production, because it was so much better than Windows 3.1 -- even for Windows 3.1 apps.
By 1995, I tried to get people to use NT 3.51 in magazine production, but the GUI put them off. NT 4 fixed that, and it took off very well for everyone except cheapskates with rubbish PCs.
That doesn't hurt my feelings, I resemble that remark.
I've confused people even worse, at more inopportune times, too.
This time of course it was about Linux more so than Windows and my suggestion that orgs that have always used Windows can make a significant step toward everyday Linux by first using Linux only for the internet and continue to use Windows for everything else.
Similar to the way small offices which had no early need for networking first got one dedicated PC for web use when they decided to start seeing what the internet was about.
It didn't matter what OS the isolated PC was running, it was handled not much differently than a fax machine, it might as well have been Linux but Linux was not made for Windows people at all.
Today desktop Linux is excellent for the internet at least, so might as well make that one step toward weaning away from Windows.
Back in the 1990's around here offices which were big enough for networking to be worthwhile had already connected their DOS machines so when they got Windows they used that to continue doing what protocol they were doing.
It was all over the place, you've shown some incredible details, I know some went from DOS to 3.1, others waited until 95, then some on 3.1 skipped 95 and waited until 98.
In the mean time the internet began to get popular.
Which is when plug & play LAN really came true for the mainstream small office whether they thought they needed networking or not. They may have a word processor on every desk but networking them for the first time was not going to happen if it required the services of some kind of professional they didn't already have on the staff.
After a while everybody wanted internet on every desk always on all the time which big companies were already enjoying. So they bought a lot of LAN cards, routers and cable and plugged & played. Sometimes needing professional help to get it sorted out afterward.
But we can be fairly old-fashioned and lots of our offices around the world still had a communication room where a DOS PC was in email service before any internet, and it was upgraded to Windows later. This was in there with the even more vintage facsimile unit and of course the traditional Telex machine for which the room was originally built. For some, the only suitable PC was a rubbish PC too, to add to the mix :\
But there's just something about communicating with the world that's a special condititon fundamentally.
I don't claim to be that accurate on networking details, I didn't deploy it much in my office until I could plug & play myself.
But for a while there I did have an old W3.1 box downloading MP3s even though it was not powerful enough to play them.
And experimented with W95's just not TCP/IP, didn't deploy that just wanted to see what people were encountering at the time.
I do remember it was the arrival of the 98 version combined with the dominance of TCP/IP, which first brought us to where it's like nobody's ever dreamed of a concept where all Windows PC's aren't on the internet at all times.
I know a little more but I just didn't want to go there since I'm no expert. My message was too long already, and I didn't even think very many people would actually try using only Linux when accessing the web from an otherwise Windows PC. But it works for me not much differently than having a separate communication room, only more convenient.
I hope my apology can be accepted for leading to such misinterpretation.
I recognized you were reading my message quite literally, but it was good to have a more detailed account of these vintage conditions from an experienced early adopter.
Quite a contrast to the vast number of small offices which only jumped on the bandwagon when it really took off, without any professional help at all.
It's worth it, it makes everyone less misguided.
Then it goes weird (IMHO) though...
> my suggestion that orgs that have always used Windows can make a significant step toward everyday Linux by first using Linux only for the internet and continue to use Windows for everything else. <
Do you mean now or in ~1998?
Either way it's a bit odd.
Now -- who uses computers without using the Internet? Then -- well, it was possible -- I built an Internet café around Caldera OpenLinux in about 1999 -- but it was quite hard.
> Linux was not made for Windows people at all. <
I think it was. The influence of Win9x on Linux is very large. Linux desktops after Win9x barely resemble those from before it.
> Which is when plug & play LAN <
The PnP LAN? PnP wasn't really anything to do with LANs.
As far as networking not needing configuration, this was a lot *easier* _before_ TCP/IP than after. Network protocols on the LAN made networking much, MUCH harder.
> They may have a word processor on every desk but networking them for the first time was not going to happen if it required the services of some kind of professional they didn't already have on the staff. <
It totally did, you know. This was my main income source in the late 1990s.
> So they bought a lot of LAN cards, routers and cable and plugged & played. <
You persist with the misguided idea that was the reason I commented to you and then wrote the blog post.
Internet use did not noticeably drive LAN adoption. Internet adoption lagged LAN adoption by a decade or so; several entire billion-dollar companies, such as 3Com and Novell, were entirely built on pre-Internet LANs and the Internet largely destroyed them.
> But we can be fairly old-fashioned and lots of our offices around the world still had a communication room where a DOS PC was in email service before any internet, <
Lots? I have literally never ever seen that.
> I do remember it was the arrival of the 98 version combined with the dominance of TCP/IP <
Maybe for you. Not in general, I guarantee you. Again: whole point of my comment.
I mean, look, I appreciate your graciousness about this, but what mystifies me is that it seems to have had no impact at all on you and you are replying to me, saying "hey that was me, thanks for the info" and then repeating exactly the same misapprehensions I was trying to correct.
I appreciate that you may be merely recounting your own experiences, but what I was trying to get across was that they were not general and the bigger picture actually is the reverse of what you were saying.
And as other commenters here have said: Win98 had almost no significance on this at all.
Windows for Workgroups: big deal. Windows 95: bigger deal. NT: huge deal. Windows 98: a minor point release and nothing more.
I guess I've got friends in low places :)
You have really given a whole clearer picture of how things on the large scale evolved into what we have today.
Now on the internet I really do use Linux almost exclusively, even on Windows PC's which are otherwise everyday office machines. Quickly reboot to Linux or have two small PC's on a desk when there are already two monitiors anyway. And I really do plug the RJ-45 at each workstation into the internet mainly for deliberate communication sessions, and most of the Windows installations never are exposed to the web at all.
I have not felt Windows is that suitable for everyday internet use for a while now.
This was a gradual way to get accustomed to Linux being good for a lot of things besides just the internet too. Once you're using it every day to begin with.
Not in the 1990s there weren't, no – because USB sticks didn't exist yet. The first patent was filed in 2000. https://en.wikipedia.org/wiki/USB_flash_drive#History
> and most of the Windows installations never are exposed to the web at all.
So you don't update them ever? That is not a good plan.
Certainly Windows is more vulnerable to malware, but it's not that bad. I rarely use it at all, but with some current antivirus installed, a 3rd party browser (that is, not IE) and so long as you don't download anything from dodgy sites, it's fine.
As far as Linux on the WWW goes, I've been using it for this since the late 19980s with beta versions of the Mozilla Internet Suite. Nothing new here.
Colloquially known as “the public.”
Win98 was intended for, and was a good fit for, home users. It was aimed at games, multimedia, portables (not new by then, but affordable mass-market ones were quite new) and causal internet use, which for most people then meant dial-up internet. It had no firewall, for example, and so would have been owned in seconds if attached to an always-on connection.
NT 4 was intended for, and was a good fit for, business users, especially with relatively high-end well-specced desktops because it had next-to-no support for power management, no USB support, no hot-plug of PCMCIA cards, etc.
I don't think home vs. business maps well onto public vs., um, not public... do you?
I mean, business means work, for many it means offices, and most working people are members of the public, aren't they? Even if they work in state or government departments, most of them are members of the public.
"Not members of the public" to me implies, I don't know, police officers, people in the military, maybe elected officials.
Your statement NT 4 fixed that, and it took off very well for everyone except cheapskates with rubbish PCs was counterfactual, but for the reason that those were the PCs businesses liked and wanted.
I know nothing about your experience, but I know about mine, as both a full-time sysadmin and tech support guy, and later a freelance consultant, for pretty much all of the 1990s.
What you are claiming as universal truths are not.
In your world, where you saw, maybe there was little to no NT. But that is not the case everywhere. I lived and worked in London, and as a freelance consultant, I mostly worked for relatively prosperous individuals and companies, because if someone's not well off financially they don't hire consultants.
Serious businesses, in more or less every sector I worked in -- in real estate, in finance, in publishing, in media, in law -- used NT.
NT 3.x was niche. NT 4 wasn't. It did well and it didn't only take over servers; I put in maybe hundreds of NT 4 boxes for dozens of companies.
It was not bulletproof -- but it was very nearly so compared to a toy gamers' OS like Win98, and by the time W98 came out, 2 years after NT 4, NT 4 was everywhere for anyone who could afford to buy a PC specced for what they needed.
The only thing NT 4 was really bad at was laptops, and some determined folk ran it there as well.
USB stayed niche until after the iMac, so the fact that NT 4 didn't support it didn't matter to most people. What mattered is if you needed your PC to be reliable, and most business people do and will pay for a more reliable box.
I am saying you are wrong when you say that its market share was a rounding error. Not in the UK, not in London where I was from 1991 to 2014, *no*. NT 4 owned business computing. Win98 owned homes and gamers but I had a grand total of one (1) client _ever_ who ran all Win98 on their client PCs, and their server still ran NT 4.
The healthcare providers usually ran a mixture of WfW 3.11 and OS/2 on clients. Many of these clients were at nursing stations where the OS/2 machines had replaced 3270 terminals and were used strictly as terminals. When Windows 95 and Windows 98 were released they replaced WfW 3.11 and at the healthcare providers also OS/2 on clients.
I had one customer that tested NT 3.1 on a couple of servers but ultimately passed on deploying in production. With NT 3.51 deployment on certain application servers started to happen and with NT4 it pretty much replaced Netware on file/print servers and was deployed for Exchange which quickly upon release started taking large market share from other mail solutions. The only deployment of NT prior to Windows 2000 on clients I saw was system administrator PCs. Windows 2000 was deployed more widely but mostly to those doing specific types of tasks. Usually replacing unix workstations. Normal client PCs only moved to NT with XP.
FWIW: The Win95 Plus Pack didn't add Active Desktop. In fact it added IE _at all_ -- it was not bundled with Win 95 at release.
It was IE4 that added Active Desktop, to all current releases of Windows at the time: 95, 95 OSR2, and NT 4.
It was bundled with 95 OSR 2.5, also known as Windows 95 C. https://archive.org/details/X03-56247
Hanging around in vintage-computer collector fora, where now people want "retro gaming rigs" based on late-'90s kit, I am often surprised at how people's memories of this stuff have faded...
But then, it was my bread and butter at the time. Making a living depended on knowing this stuff and being able to fix it.
First one with native windows networking rather than DOS redirected drivers.
You make a good point.
I've not seen that before myself -- I mainly inhabit the Linux/macOS space now -- but the person whom I was paraphrasing certainly seemed to think, as you say, that W98 was the big deal and marked the beginning of integrated networking in Windows. Very bizarre, because as you say, it was a minor bugfix release.
There were approximately four (or six, depending on how you count) later revisions, but the big one was "OEM Service Release 2", which added support for FAT32, Infrared data, UDMA, IRQ Steering, Firewire, MMX, and P6 support. A later minor patch also added support for USB and AGP.
The later OSR "2.5" also came bundled with IE4, which overhauled the Explorer interface, adding (amongst other things) the quick launch bar, IE integration (including things like the "address" bar in Explorer windows), and the Active Desktop.
Today, many people assume these things came along with Windows 98 first, which is not true.
> Anyway: that was the sole significant improvement in networking in Win98 that I am aware of. The other thing 98 could do was drive multiple monitors, if you had the right — that is, from an extremely limited selection — graphics cards. Otherwise, it was just 95B with more drivers and the odious Active Desktop built in.
Looking back I now see that the Windows 98's "Active Desktop" was the first "builtin bloat" to the Windows Family. That was the first step that started the long march to current hated things like Cortana.
True enough, but I slightly disagree.
I had an account on it and I used the MSN. It was pretty good for its time, and streets ahead of AOL or CompuServe. Eventually it devolved into being just an instant messenger and a news website, and now, the instant messenger is gone.
Secondly, if you didn't use MSN, you could just totally ignore it and it imposed no system load. In that way, it's comparable to Windows Movie Maker in XP: totally pointless if you don't need that, and you can't remove it or choose not to install it, but doesn't actually do anything at all and can be ignored.
Whereas Active Desktop could not be ignored or removed (except by doing a custom install with 98Lite, which is why I linked to it) and it was necessary to use it to use the OS at all.
I agree. Um, perhaps I should have used a less obscure word than "odious"? :-)
"arousing or deserving hatred or repugnance; hateful; an odious crime; a false and odious comparison."
This does neglect the reason, which was well-known at the time and may be forgotten by now. MS was subject to a monopoly lawsuit by the US Department of Justice back then, and was trying to show that IE was an inextricable part of the Windows OS.
If IE were just a web browser, you could remove it and the OS would still function just fine; but if it's also the desktop, you can't uninstall it without crippling the OS -- therefore, it's part of the OS and MS couldn't be told to remove it.
And that's the only real reason.
The slight side-note being that "Active Desktop" was multithreaded, like the MacOS 8 Finder.
In Win95 or MacOS 7, if you started a file copy (for example), the whole desktop froze until the copy was done. You couldn't do anything else except sit there and watch the progress bar.
In MacOS 8, the Finder (the program that drew the desktop) got multi-threading along with lots of other improvements (desktop drawers, an easily-customisable Apple menu, and more). Start a big file operation and you could go do something else while it sat there working away in the background.
IE4's Active Desktop brought this to Win95 OSR3, Win98, and NT 4 if you installed IE4. But it was the only improvement. The rest was, as you say, bloat.
In Win95, the copy happened on the destination folder's thread. You could still open other windows and keep browsing around, but that specific window would be unresponsive.
The OS was still running; you could switch to an app and keep working, for instance. But Explorer had only 1 thread, AFAICR, and a large file operation visibly locked that thread.
I wrote about this in PC Pro in about 1995-1996. Because it had a true pre-emptive kernel, it could do other stuff while waiting for disk I/O, and on NT, enabling DMA made a big difference. As I recall, on 98, it just made disk access a bit quicker, that was all.
So there was someone. How about that?
BTW & FWIW, I'd say "bloat" means something that you don't need that adds complexity, memory overhead, and attack surface to a product, and probably makes it slower, use more disk, and generally makes it heavier and less responsive.
I don't think desktop wallpaper counts as any of that. The memory usage is trivial, it needs no CPU or disk space, and it has no UI after it's been set.
It's chrome: unnecessary functionality added only to improve the visual appearance -- but in my definitions it's not bloat.
A bitmap of 1024x768x24 is around 2MB uncompressed in memory, which could be a lot at the time. (I remember having 24MB on my machine, maybe a bit outdated for W98?) However, I guess that when a pattern is used as wallpaper, memory usage is indeed trivial (only store the pattern, blit repeatedly)
I think it got swapped out as soon as it could be, though. I remember moving windows aside on low-end machines and hearing the hard disk chug as it loaded the previously-obscured bits of the desktop image back in, scaled them and redrew them.
(I don't remember when Windows gained the ability to scale the wallpaper. IIRC, it was in the Windows 95 PowerToys – the base OS couldn't do it. I think it also only understood `.BMP` files originally. JPG and other compressed formats only came later – probably in 98, come to think.
I had a few such machines that could do 1024×768×24, but they were pro-grade kit, and as such, I ran NT 4 on them, not GameOS®™.
About 1997 or so I was rather proud of my Pentium II 450 MHz in a Baby-AT case (not ATX) which had 2 Matrox cards, allowing me to run a dual-head desktop at 2048×768. I think I then went from 2×14" screens to 2×15" which allowed me to run twice 1152×864: 2304×864. So… much… desktop!
The selection included all common S3 models, and you just had to have one S3 graphic card to be compatible, and they were already dirt cheap. It was AMAZING.
This was the real killer feature (most of the time I didn't even have one IP address because dial-up internet was expensive).
Oh cool. Glad to hear it provided entertainment. :-)
No, really. I have a day-job as a writer and if I thought I could have crow-barred that in there somehow, I'd have done so. But I could not find a citation for the 4-IP address thing. So, the result was purely for fun.
As so many tcp/ip, we start to take courses on a new company called Cisco and it’s competitor. Sun, hp and we opt for aix. Synapse/power build but we use oracle, we opt for the disaster oracle form 4, … My master degree with term paper on ospf. It was all 1990s craziness.
The only thing is that the network driver of windows is no good in win95. There are three and we used a place in Australia start with W. I can’t remember win98 helped that much.
Far East perhaps.
Yes home start struggle with 1200/2400 modem. But lease line is not that fast just more stable. And bi-directional.
You are not sure about which bit?
Lotus Notes became IBM. Nothing to do with Novell. Novell's equivalent was Groupwise. I used it until 2018. It was... not great.
But it didn't depend on TCP/IP. I don't think Notes depended upon TCP/IP until Domino came along in a much later version (5? 6? No idea. Last used it in 2007 and didn't like it then.)
That said, agreed - LJ is trash.
Hijacking the back button is never a good idea...
It never went away. I've had my main blog there since 2002 and tech one since 2006. It works, it's free. I don't have a paid account and never have, and I do not see the problems people mention because I have a working ad-blocker.
I am sorry for those who don't.
Microsoft published the Winsock Spec/API up until 1997, when their stack finally got good enough to make the other apps obsolete. I think they still use the Winsock API to this day.