47 Comments
Dec 5, 2023·edited Dec 5, 2023Liked by Babbage

It's worth noting that MS supported the Alpha processor. I remember that NT setups came on two distinct folders: Intel and Alpha.

Expand full comment
Dec 4, 2023Liked by Babbage

Categories of devices that are doomed.

1. Laptop computers. These will be replaced by interfaces that are driven by a personal compute device and work compute devices.

2. Desktop Computers. These will be replaced by interfaces that are driven by work compute devices, and personal compute devices.

Compute device will continue to shrink. The reality is all interfaces can be driven by the compute that is carried by a standard human today, and that capability will only continue to exponentially increase.

Expand full comment
author

Hi Alan, 100% agree on the desktop which is presumably kept alive now by gamers? Not so sure about the laptop. I have an iPad and a keyboard is attached to it >75% of the time so it really wants to be a laptop!

Expand full comment

I prefer my desktop for work, let alone gaming.

Expand full comment

Even there - the desktop is being replaced by gaming laptops and the newer 4x4 processor boards (I have a NUC12 under my desk that replaced a 19" tower and is MUCH more powerful while being only 8.5 x 2.5 x 7.5 inches in size).

Expand full comment

There is one mini architecture that (kind of) lives on to this day. The TI-990 mini became the TMS9900 microprocessor, and the architecture then found its way into millions of MSP430 microcontrollers.

So though the mini form factor has kind of* disappeared there's at least that one case of a mini architecture surviving into the present - even if it's ridiculous that it's powering things like the control system of an electric toothbrush.

* Arguably the form factor also lives on in 'size of a refrigerator' rack scale systems like VxRack or Oxide.

Expand full comment
author

Thanks Chris. I was aware of the TMS9900's mini roots but not that it had made its way into microcontrollers. Must have a look as I've always been intrigued by TI's designs.

Interesting challenge on Oxide etc. I suppose I was thinking that these would typically wouldn't be deployed in isolation but happy to be corrected.

Expand full comment

The 430 family is a set of lovely tiny microcontrollers that come in all sized and prices. The TI dev tools, however, I'm not so fond of.

Expand full comment

They had their own OS, compilers, utilities, etc.

No windows, no common unix.

Restricted peripherals, memory, comm

PC eventually beat them out on all counts including maintenance costs.

Expand full comment

Oh, and VMS had DECWindows!

Expand full comment

VMS is still alive (won't say "and well").

Expand full comment

Brief time at IBM? Do tell! (yes, I read the footnotes)

this is making me want to pay a repeat visit to the Computer History Museum in Mountain View, btw

Expand full comment
author

Marketing support in the mid 1980s, trying to help IBMers sell PCs (very easy) or System/36 (impossible). One anecdote: after s S/36 demo to a room of customers, which seemed to go well, one of our team asked if there were any questions. Complete silence for a moment until a hand went up. 'If it's this slow with one user, how slow is it with five?' Room dissolves into laughter.

Do share in one of your letters if you go to the CHM!

Expand full comment
Dec 3, 2023Liked by Babbage

I think pinning failure of that era to CISC is very difficult. DEC also had two operating systems that failed, and as you wrote a RISC chip that didn’t last. It’s true for other OSes like Be, OS/2, Amiga, etc., and it was true for other RISC chips, most notably the POWERpc.

My point is that DEC was much more than the VAX cpu, and many others failed in those other arenas too. x86, Windows, Office, the technologies and the marketing were a very special and well-controlled combination that left many companies in its wake.

Expand full comment
author

Hi Chris, Thanks for commenting. I tried really hard - and obviously failed! - to avoid drawing a straight line from VAX's CISCy ness to DEC's failure. Completely agree that DEC was much more than VAX and that Wintel was a steamroller that rolled over many companies. I think my key point on DEC was that its origins as a mini company made it hard to pivot to being something else, even with great technology. Probably needs a much longer post to explore properly!

Expand full comment

I had a PDP 11/x, 2 MicroVAXes, and finally an AlphaServer 4100.

Costs were a huge factor in the demise. I had probably the first 1TB of data in MA in a DEC RAID cabinent full of hot swappable 40GB hard drives, and I spent 100k or so on it.

Add in service contracts for 24x7x365 DEC field service, the costs of software licensing, etc, and the cost per user were huge.

Still, VMS is a better OS than most today, and I would give anything to have a PC version of Datatrieve vs some of the stuff today.

Expand full comment

Doesn't a VMS machine have the world record for uptime?

Expand full comment

Heh, I never looked.

DEC Field service was amazing though. The one time I had a major outage and nothing was working, my parents (this was a family business) were in Germany, and I needed to be up online on Monday morning. Oh yah, its the weekend before Christmas.

I had two guys driving parts between Woburn and Feeding Hills MA and spending hours upon hours rebuilding the machine to get it up and going.

Quite amazing. I also convinced my dad I was the guy to run the company from that point forward.

Expand full comment

I worked with a lot of DEC mini's from the mid-80's until the mid-90's. Ended up doing XWindow/Motif dev for DEC RISC and DEC Alpha. I started as an Ultrix SysAdmin.

Expand full comment
Dec 4, 2023·edited Dec 4, 2023

I ended up selling that company in 2005 but stayed in the same field of credit reporting, but now in Compliance issues and legislative affairs.

Early 20s for that company and learned how to build nearly everything and program because it simply wasn't easily available yet. Cat5 cable in 1000 foot spools with crimping tools. We got lucky as a small family biz to find a remote programmer in the early 90s who knew databases and could build search algorithms before data science was a thing. :)

Expand full comment

A lot of VMS still exists in the Windows kernel.

Expand full comment

VMS? I thought it was Pink that Dave Cutler used as the model for NT.

Expand full comment

Workstations are still around and have for the most part absorbed the useful aspects of the mini computer, for example a thread ripper pro Workstation like:

Edge AI – EDAI-5975WXQR6 AI Workstation

$47,850.00

AMD Threadripper PRO 5975WX

256GB RAM

4x NVIDIA RTX A6000

2TB PCI Express SSD, 6x 2.5" 2TB SSDs

Expand full comment

"Indeed, DEC would put considerable effort into developing RISC architectures rather than putting that development effort into VAX."

VAX 9000 says "hold my beer". (arguably, DEC never recovered from that debacle; they literally started with the still-smoking remains of the Trilogy Systems moonshot)

Expand full comment

Worked for a company that "skipped" the transition cycle from proprietary Sperry 77, Sperry 620s upgrade because the division was being bought/sold/becoming an independent company.

Arguably the upgrade would have been Mini workstation VMS servers. When they finally did the transition.

PCs 486 architecture was advance enough & cheap to perform the job.

Just wait the cycle will began again just like when Ross Perot use to park a trailer filled with digital equipment performing payroll services for companies. Someone using a quantum computer the size of a truck, will perform Material Science, Medical research, Extreme cases of Structural integrity, etc. for companies willing to pay.

Expand full comment

And MIPS, too. NT was originally developed on MIPS-based machines.

Expand full comment

Evocative piece. Brought back early 80's memories of my spell with a Microdata (McDonnell Douglas) Reality, and its Richard Pick OS.

Expand full comment

No, that wasn't likely the case, it was primarily the case that specially the sales and marketing folks at the "mini-makers" for far too long thought that only with big computers, you could make big money. And then we're left in the dust cloud when the PC and networking market exploded. I know, as I was working at that time for a company that was still trying to sell DG Nova clones and jumped ship well before they went under.

Expand full comment
Dec 7, 2023·edited Dec 7, 2023

Sure, the TI-990 spawned the TMS9900 family (similar to the way the PDP-11 spawned the F-11 and J-11 parts; T-11 doesn't count here), but there's only superficial architectural similarity between the MSP430 and TMS9900. In particular, the MSP430 lacks the fast-context switch of the 9900. It's no less accurate (and no more useful :-)) to suggest the MSP430 was influenced by the PDP-11.

In particular, Texas Instruments themselves describe the MSP430 as the descendant of the TSS400, a relatively obscure 8-bit MCU with on-chip interpreter for a macro language, completely unlike the 990/9900/99000 families (https://www.ti.com/sc/data/msp/databook/chp8.pdf).

Also, in chapter 1 of that databook, they mention the PDP-11 in glowing terms:

"The 27 core instructions combined with these special features make it easy to program the MSP430 in assembler or in C, and provide exceptional flexibility and functionality. For example, even with a relatively low instruction count of 27, the MSP430 is capable of emulating almost the complete instruction set of the legendary DEC PDP-11."

I think they just re-used mnemonics as companies are wont to do .

Expand full comment

It's true, DEC would have put a lot of effort into developing the RISC architecture rather than putting that development effort into the VAX. When it became clear that RISC architectures were a good idea, then the days of VAX as an architecture were numbered. This also had commercial advantages which is large because it has the scale to support large investments, both in architectural development and the manufacturing process to build it. You may protest that x86 is a CISC architecture. For clearer information, you can visit our website https://furloughedfoodieslondon.co.uk/

Expand full comment

It seems the author of that article was born in the 90s, maybe, and never actually worked with a mini computer. Or the question why there are no minicomputers anymore wouldn't present itself to begin with.

Two simple reasons: For one, the computing power of "micro-computers" increased dramatically in the early '80s that they reached the same processing power than those estate fridge sized boxes. And on top of that, PC networking evolved dramatically and LAN setup simply eliminated the use for multi-user machines connected via serial terminals....

Expand full comment

It's fair to point out the makers of minicomputers were aggressively shrinking them into microprocessors throughout the 1980s. Perhaps the only reason the minicomputer form-factor persisted as long as it did was a desire to make it physically difficult to steal the company computer.

Expand full comment

I might call Sun SPARCStations minis despite their desktop size. It’s their relationship to earlier OSes that ran on Vaxen that defines the category in my mind, not their size.

Expand full comment

SPARCstations (and the similar products at the time) were workstations, arguably means something like "the power of a minicomputer + framebuffer", and there were minicomputer-sized/featured products with those microprocessors, by the '90s we called them servers. I seriously think the datacenter people saw it as an advantage that one person could not carry off the minicomputer.

Expand full comment

Yes.

RISC Challenge server?

The Sun 450. It’s been so long, I don’t remember the names as well. I’m thinking of up-scaled workstations as servers like you say that had cabinets about 19 in. wide, really deep and three feet high.

Expand full comment

Another reason why minicomputers vanished, instead of evolving (as they should have) into the low-cost market segment, is the willingness of businesses, including [very] many who should have known [much] better, to run their business on consumer-grade toys. Bill Gates's defining insight was that financialized business does not need the right answers: they just need to look good.

Expand full comment

Just found a non-consumer grade toy's story: https://www.youtube.com/watch?v=41Gv-zzICIQ

Expand full comment

"on consumer-grade toys" - just imagine how cost prohibitive would be to run a company on minis/mainframes on the current level of general computer use. Then imagine how much you should wait for a new browser, [insert your favorite language] compiler/framework, ...

Expand full comment

The cost -- absolute cost and price/performance -- of (say) MicroVAX/Alpha would have adapted to a market in which buyers understood that it is not possible to productionize toy hardware or toy software. My employer was one of the very first -- pre-release -- adopters of 32-bit Windows on the server side; all of the imagined economies proved illusory, and buying the cheapest kind killed the company very quickly: the software Simply. Did. Not. Work. (multiprocessor support, RPC). Once a toy, always a toy. But on the demand side, it was all about pretending.

Expand full comment

... and they went tits up while their competitors, who remained on VAX, are still flourishing ...

... and we have not mentioned Novell Netware yet ...

Windows Server came 10 years after MicroVAX, and instantly removed hardware vendor lock-in and their premium prices, too.

Reliability?

Have you seen a Novell Netware 4 reporting that it lost it's system partition (due to some ISDN installer men) but running flawlessly till the end of the day with 15 users?

Have you seen a Windows 2003 Server reporting that it lost CPU1 (due to a Xeon power module failure) but running till the replacement arrived?

Have you seen a puny Windows 10 Home reporting that it disables Core 2 because it's L2 cache is unreliable?

Expand full comment

I'm not sure what you are trying to say. Undeniably, Windows has gotten better, and since Windows 7 it has actually been enjoyable to use -- <b><i>as a consumer</i></b>. But my then-employer, whose identity you have probably guessed, bet the farm on what Microsoft told them 1991 -- 1993 and much of it was false. They were not totally heedless: they paid for a source license (the second in the world; nobody can get one today for any amount of money), they sent many people to Redmond many times. But they missed the fact that the "multiprocessor" kernel did not actually support multiprocessor device drivers, and they missed the fact that the call-cancel API in RPC was a stub that always returned true. The latter "issue" burned over a billion dollars of shareholder value in a matter of weeks. (I had the privilege of confronting the applicable Microsoft engineer about that in 2000, long after it had been fixed and none of it mattered any more.)

But when I say "toy", I am not confining that description to NT 3.1/3.5/3.51, nor yet to the entire Windows software ecosystem, but also including the hardware ecosystem. They kitted out an entire datacenter with 2U dual-Pentiums (90 Mhz...!) that turned out to have bad RAM. (The fact that there was no way to write MP-safe kernel code surfaced as a side-effect of the forensics on that.) That same new datacenter was also stuffed full of routers (from a company who have since died) with a firmware bug such that 90% of attempts to establish a TCP connection failed. None of that was Microsoft's fault, but none of it would have happened if the CTO and his people had stopped for a split second to ask, <b><i>why are we buying toys</i></b>?

Your comment about hardware lock-in is merely amusing, inasmuch as the tooling and documentation in the Windows DDK was thumpingly useless until very, very recently. A tiny handful of vendors bit the bullet and paid Microsoft for the good doco and the good handholding; everyone else had to build towers of guesswork. It is getting better, but buggy device drivers are still a problem.

Expand full comment

Oh, yes, 4-way MP system that was able to use only one processor... for even writing to a HDD... amusing.

Yes, device driver development requires knowledge[1], but even now it is the case. VAX was like iOS now, you had to develop "one" device driver - but this change washed away the ivory towers with vendor lock-ins, so nobody looked back since then. "Server for the masses" worked.

Looks like your employer learned a lot in the VAX world if they sucked this huge way in HW, SW and network at the same time :) Sure, everything was toy level there - but who signed the acceptance tests?

[1] https://www.geoffchappell.com/studies/windows/km/ntoskrnl/source/tree/310.htm

Expand full comment