Kevin,

I found two articles that gave more technical information on the exact problem. First of all is this issue of the chip that was used to supply "serial control" of various devices. Many people might not think about it, but the keyboard and mice are also "serial devices" and used up two ports of the four port DC7085 gate array:

These systems have four asynchronous serial lines that are provided by a DC7085 gate array. Of the four serial lines, only the third line has the required modem control signals to support a modem. A 4-pin MMJ connector is provided for the keyboard line, a 7-pin DIN connector for mouse line, and two 6-pin MMJ connectors for printer and modem lines.

Now we put in who articles about what lines and pins were actually implemented:

http://www.vanade.com/~blc/DS3100/pinouts.html

https://www.tuhs.org/Usenet/comp.unix.ultrix/1991-May/005327.html

Now it appears as if the DS 3000 had two "serial ports", one designated for a modem and one designated for a printer. Reading them brings back more of the "issues".

One issue is that many Unix systems were not used just as a desktop system, but as "servers" too. A prime example of this was that a VMS Workstation was only allowed a single user to be logged on at one time. A multi-user license was not allowed for VMS Workstations through a dictum put forward by VAX product management.

This, of course, was not what Unix people (and particularly Sun customers) would do. They might have a single person using the keyboard, mouse and main screen, but have several others logged onto the system over TCP/IP.

You can now skip ahead to more about the issue of serial ports, but you will be missing an interesting part of DEC history:

For years DEC refused to sell a multi-user license to workstation users. We even had a separate product called Ultrix-32w which was both the Ultrix-32 software plus the graphical server and came with a two user license (so either uucp or the network administrator) could log in at the same time as the user at the main keyboard. However, no other general user could log in at the same time. So the base Ultrix-32w came with that two-user license, and there was no "upgrade".

Eventually I found this "rule" so damaging to Digital's Unix business that I decided to fight it. This "rule" might have made sense in the original days of 1 million instructions/second CPUs and unaccelerated graphics, but made NO sense with the much faster and accelerated graphics. We were being whipped senseless by Sun's marketing team.

So I traded this "rule" down, and finally found that it had been generated by Jesse Lipcon, at that time a VP of workstations. I called him up and explained the situation. "Oh, I NEVER meant that to apply to UNIX Workstations...only VMS Workstations" said Jesse. "The Unix market is much different."

So I held a meeting the next week. I picked the largest meeting room, expecting there would be much argument. I made a huge presentation about the whole issue and how much business we were missing. I pointed out that it was really up to the customer how many people or processes would be on that system, and DEC should not be limiting it artificially.

The day of the meeting, the room was packed. There were people standing in the hallways.

I stood up at the front, made the presentation, and asked if anyone had objections. No one raised their hand.

"Then why are you here?" I asked.

"We just came to see the fireworks." was the answer.

"Then Digital now sells multi-user workstations." I declared.

Of course things were a little more complicated than that, since there were changes to documentation, licensing, and a whole project around adding another graphics tube, keyboard and mouse to the hardware (the "Multi-User Caylith" project), but we charged ahead with the multi-user system.

Back to the serial ports:

So the VMS viewpoint of workstations you might have a printer and a modem hooked to the workstation, but more likely these would be attached to a server someplace to be shared with other workstation users.

From a Unix standpoint, your workstation might ALSO be a server, and might have a printer and modem OR TWO MODEMS attached to the workstation,

On either serial port they worked fairly well at lower transfer speeds with handshaking, but with higher speeds the handshaking tended to break down, especially if you hooked a second modem up to the "printer port".

Another issue was the lack of the "CD" line, that tells the system the modem has lost connection, and requires that the software on the system recognize this through a timeout.

This violated the concept of "DECstandard 052", a standard that was developed for modems hooked up to DEC's systems. Before the adoption of "DECstandard 052" in order to hook a modem up to a computer and a country's telephone lines you had to test every modem with every piece of hardware and every operating system in every country. This was not only a very expensive process, but took a long time after operating system releases.

By developing DECstandard 052 the company made this a lot easier. The hardware and operating system groups only had to certify against DECstandard 052, and customers could then connect up any "standard" modems and know that they had the modem control signals to reliably interact with the modems of the country.

The DECstation 3100 did not meet DECstandard 052:

DEC STD 052-0 Operational Requirements for Serial Terminals and Computer Interfaces Operating as DTEs Connected to EIA RS-232-C or CCITT V.28 Modems - Part: EL-00052-00 - 1981

We've run both UUCP and SLIP on 3100s here; the modem-control "problems" really aren't that serious a problem unless you are a high-security sort of place. What is missing is the CD line that lets the modem tell the computer that it has lost its connection. This means that when the other end (or the phone company ;-) decides to hang up on you, the modem knows about it right away, but the software has to use a timeout mechanism, because the modem's attempt to say "Hey, we just lost the connection" does not get through.

This allowed another call to arrive at the same port, and perhaps gain control of it without having to authenticate.

Unlikely? Yes. Impossible? No.

Using the printer port for a modem made the issue even worse. After all, you did not expect the printer to talk back to you at high speed.

So now you know the REST of the story.

On Sun, Jul 7, 2024 at 1:32 PM Kevin Stratton <kstratton@fastmail.us> wrote:

I found this information interesting.

The statement

"To save a couple of pennies per system the
> VAX hardware designers determined that having full hardware modem
> control (DECstandard 52) on the board was not needed, because "very
> few workstation users hook a modem up to their serial line".

has two different meanings (to me) in the context of this discussion. They are:

1) RI and DCD are not implemented in hardware (Ring indicator and data carrier detect)

2) A subset of RTS, CTS, DSR, DTR as well as RI and DCD are not implemented in hardware.

Does anyone want to clarify this in greater detail? Just curious.

On 7/6/2024 8:59 PM, Jon "maddog" Hall via WLUG wrote:
John:

That's an interesting statement. I would think it was the same case
and port setup, but a different motherboard since you couldn't just
slot a MIPS processor into a VAXstation processor.

That is why I said "based".   "Based" means that you did not have to requalify a lot of the parts, that you could use a lot of the same Bill of Materials....it does not mean you use exactly the same PCB.   In the case of the VAXstation/DECstation 3100 that meant we used the same serial io chips and did not provide the hardware flow control.

Yes! It's coming back to me, the 3100s were horrible for serial,
though I think they had fixed it for the 5000 series systems, since
the DECstation 5000 and the DECserver 5100 were identical, except the
DECStation had a graphics card and cost 10-20% _less_ than the
DECstation 5000. So we bought DECStations and used them as servers.

The server groups were different engineers, and the 5000s were designed after the DECstation 3100/2100 series.

Actually we MIGHT have caught the issue with the serial port for the DECstation 3100/2100 because I brought up the issue in a hardware meeting and was assured that it had been fixed because (and I can still hear this person's voice in my head) the DECstation was a REAL UNIX MACHINE.   I should have found that person later and beaten him to death.

David:

Actually, Cutler was already in trouble with management because of the Prism project. Prism was the first try at building a RISC machine (this was in the 1985-1986 time frame) and it was managed by Dave Cutler. Unfortunately, Dave was not a good administrator. He was doing a very bad job of managing the project and doing a very bad job of keeping management up to date on what was going on. Finally, management cancelled Prism and that is when they decided to go with the MIPS architecture.

That is not what I experienced. I will not comment on Dave's skills as an administrator or manager of the project. I do know that he had the very ambitious project of doing the next-generation architecture and the next generation operating system. He wanted a micro-kernel based system with "personalities", one personality being VMS and the other personality being Unix.    One issue he had was when there was a clash between personalities he would always favor VMS and therefore the Unix personality would not be "perfect". I felt this was a mistake, since Unix people did not tolerate things that did not smell like Unix. VMS people could have been led over to a new environment over time since DEC controlled the path with VMS.

The reason I know this is that I personally interviewed with David Cutler for the position of Unix Product Manager for his new operating system.    The interview was going fine until I tried to explain to Dave why his strategy was wrong and he had to make the Unix side really compatible.   Then things got ugly, as David really did not like to be told he was wrong.

Ken did have the bad habit of cutting off Dave's development money, then giving it back to him again.   I seem to remember that Dave was still going full steam ahead while the PMAX was being developed, which is one reason why he was caught so flat-footed when PMAX was announced. It was kept REALLY SECRET.   It probably did not help that Ken reportedly asked David in a gleeful tone if the new chip would be as fast as the PMAX....

"Will it be THIS FAST, Dave?" was the reported question.

It is also my belief that the Alpha chip came directly from Dave's work on the PRISM, and that Windows NT came directly from David's work on the microkernel with personalities.    DEC never "decided to go with MIPS" except as a stop-gap measure for Unix.    If DEC had "decided to go with MIPS" they would have ported VMS to it, and since that never happened, it was obvious that DEC had not embraced MIPS.

md

On Sat, Jul 6, 2024 at 3:27 PM John Stoffel <john@stoffel.org> wrote:

>>>>> "Jon" == Jon \"maddog\" Hall via WLUG <wlug@lists.wlug.org> writes:

> The DECstation 3100 (aka the PMAX) was a system based on the same
> motherboard as the VAXstation 3100, except with a MIPS processor.
> It was the first time that Digital had built a system using another
> company's chip architecture.

That's an interesting statement. I would think it was the same case
and port setup, but a different motherboard since you couldn't just
slot a MIPS processor into a VAXstation processor.

> We had a choice between supporting the MIPS R2000 chip as "big
> endian" or "little endian".   We chose to support "little endian" to
> be compatible with PDP-11s, VAXen and Intel chips, rather than be
> compatible with other manufacturers. This, in my mind, was very
> important since it let us share data over NFS and give greater
> compatibility with programs and libraries of applications already
> running on other Unix-like systems such as Ultrix, BSD, etc.   Sun
> experienced this grief when they ported Solaris off the big endian
> Motorola and SPARC chips to little endian Intel.

> One of the few issues of the DECstation 3100 (and 2100, known as the
> "PMIN" after that) was that the serial I/O port did not support
> hardware handshaking.   To save a couple of pennies per system the
> VAX hardware designers determined that having full hardware modem
> control (DECstandard 52) on the board was not needed, because "very
> few workstation users hook a modem up to their serial line".   What
> they meant was that "very few VMS workstation users hooked a modem
> up to their serial line", opting for DECnet to handle their
> networking.   Therefore the pins necessary to support hardware modem
> control were missing from the DECstation product lines.

Yes! It's coming back to me, the 3100s were horrible for serial,
though I think they had fixed it for the 5000 series systems, since
the DECstation 5000 and the DECserver 5100 were identical, except the
DECStation had a graphics card and cost 10-20% _less_ than the
DECstation 5000. So we bought DECStations and used them as servers.
:-)

I used to live and breathe the DEC options catalog, which was
something like an inch thick book in purple (maybe blue? maybe changed
each year?) with all the various options and configurations you could
get.

> Unix users, on the other hand, often had modems hooked up to their
> workstation, to handle cu(1), uucp(1) and other serial line
> protocols.   We eventually got around this by better refining
> software modem control, but it was painful.

The EE department had Sun 1s at WPI, and they have the 9600 baud
bitnet connection running of a VAX 11/780 at the time.

> The proposal, design and release of the DECstation 3100 was a
> top-secret project inside of DEC, kept secret from even many people
> in the Digital Unix group in Nashua, New Hampshire.   The
> development of the system was done in Palo Alto's WorkStation
> Engineering group and the software engineering was done in the
> Western Software Labs in Palo Alto, albeit by taking key engineers
> from the Nashua facility and transplanting them temporarily to Palo
> Alto.

> The implementation was kept so secret that it caught Dave Cutler,
> then working on DEC's next-generation operating system and chips, by
> surprise when it debuted at a Board Meeting in Maynard,
> Massachusetts.   Shortly after that board meeting Mr. Cutler left
> Digital and moved (with some of his team) to Microsoft.

Do you think that was just the straw that broke the camel's back?

> The DECstations only were manufactured for a couple of years, as the
> DEC Alpha processor replaced MIPS in Digital's lineup.   Although
> originally promised that OSF/1 would be ported to the DECstation
> line, this port was dropped due to realizing that there would never
> be an application base to support the customers of a DECstation
> OSF/1 port.   It was better for ISVs and customers to maintain the
> Ultrix base for the DECstations than to split the base into Ultrix
> and OSF/1.

Yeah, we had a ton of DECstations and a bunch of various early model
Alphas at WPI which we beta tested for DEC. The AS2100 with 8 DEC
StorageWorks canisters (which Netapp used on their early F220, F330,
F740 NAS systems) holding 3.5" disks, dual ported, hot-swapable. Nice
gear.

The move from Ultrix to DEC OSF/1 was a bit painful, but not terrible
as I recall, but I'm sure I've blocked all the bad memories from then.

Thanks for the walk down memory lane!
John
_______________________________________________
WLUG mailing list -- wlug@lists.wlug.org
To unsubscribe send an email to wlug-leave@lists.wlug.org
Create Account: https://wlug.mailman3.com/accounts/signup/
Change Settings: https://wlug.mailman3.com/postorius/lists/wlug.lists.wlug.org/
Web Forum/Archive: https://wlug.mailman3.com/hyperkitty/list/wlug@lists.wlug.org/message/UUCPQ4AAGHN37ST6IFAL7NSACXXFI6FE/