In article <b767aa9a-e341-426e-8215-0a9e9f6697bcn@googlegroups.com>,
jimc...@gmail.com <jimcausey@gmail.com> wrote:
>On Monday, September 25, 2023 at 5:15:50 PM UTC-7, Dan Cross wrote:
>> The early days of Windows NT are well-documented in the book,
>> "Show Stopper!" by G. Pascal Zachary. In short, they used OS/2
>> and 386 machines; NT was self-hosting within a couple of years.
>
>Zachary wasn't very technical and made a number of mistakes in that book, although overall his story is well-researched and compelling. NT
>was originally brought up on single-board Intel i860 hardware, followed by MIPS DECstations and then i386 hardware; Cutler insisted that the
>team not focus on i386 because he wanted to keep NT from becoming wedded to the x86 architecture.

I knew about the i860, but this is the first I've heard about
the DECstation port (I thought they used MIPS Magnums or
something?). What was the initial development platform,
though? Certainly once it was self-hosting it could be any
supported platform, but before that? I was under the
impression that was mostly PCs running OS/2.

- Dan C.

On 10/3/2023 1:40 AM, jimc...@gmail.com wrote:
> On Sunday, September 24, 2023 at 11:04:53 AM UTC-7, John Dallman wrote:
>
>>> If DEC had went after the low end market with the C-VAX, I really
>>> feel that DEC would still be with us today.
>> Maybe. The MS-DOS hardware and software industry was already very well
>> established, and competition had driven hardware prices down a lot. DEC
>> would have had to pick some niches to target and win several of them.
>
> Agreed. By the time CVAX launched, the damage had already been done -- CVAX was not price-competitive with x86. Even Alpha wasn't really price-competitive with Pentium.
>
> The opportunity (if it ever existed) was squandered much earlier, before the first MicroVAX parts shipped, when Olsen decided to kill the strategy to sell them to OEMs and drive volume. John Mashey describes a number of ways where even that strategy might have failed, but there was zero chance of pulling it off by 1987.
>

Well, yeah, almost every time, volume wins ...

DEC would have had to sell the C-VAX really cheap, perhaps at a loss for a
while. To have a market, one must first acquire that market.

--
David Froble Tel: 724-529-0450
Dave Froble Enterprises, Inc. E-Mail: davef@tsoft-inc.com
DFE Ultralights, Inc.
170 Grimplin Road
Vanderbilt, PA 15486

On 2023-10-03 01:58, Scott Dorsey wrote:
> gah4 <gah4@u.washington.edu> wrote:
>> It is some years now, so I don't remember the details, but I am pretty
>> sure that there was one that worked even if you didn't have an account.
>
> Decnet to arpa? Sure, there were lots of them and none that I know
> required an account. It was just a polite service people provided.
> The best one was at Columbia which had really good connectivity (and also
> bitnet connectivity) so you could do "fredbox::fred@columbia.edu" as I
> recall.

For mail, yes.

That still happens... Try sending to "pondus::bqt"@mim.stupi.net and
you'll reach me on my PDP-11/93 running RSX-11M-PLUS at home.

Johnny

On 2023-10-02 17:40, Dan Cross wrote:
> In article <ufeno9$g48$2@news.misty.com>,
> Johnny Billquist <bqt@softjar.se> wrote:
>> On 2023-10-02 14:56, Dan Cross wrote:
>>> In article <ufe5e0$nbu$7@news.misty.com>,
>>> Johnny Billquist <bqt@softjar.se> wrote:
>>>> On 2023-09-29 07:07, Lars Brinkhoff wrote:
>>>>> Johnny Billquist wrote:
>>>>>>> NCP and TCP operated in parallel on the ARPANET for a while.
>>>>>> How would they interoperate? TCP and NCP are not exactly compatible in
>>>>>> any way.
>>>>>
>>>>> Details are found in the "Internet Protocol Transition Workbook".
>>>>>
>>>>>> you might have some machines that would act as gateway between the two
>>>>>> networks.
>>>>>
>>>>> That's exactly what they did.
>>>>
>>>> That's what I expect. SO we have ARPANET, which is talking NCP, and you
>>>> have hosts that talk TCP/IP that can communicate with hosts on the
>>>> ARPANET via a gateway. Does that mean the TCP/IP hosts are on ARPANET? I
>>>> would say not. Just as hosts on my hobbyst DECnet are not neccesarily on
>>>> the internet, but they can communicate with hosts on the internet when
>>>> there is some gateway in between that can forward stuff for them.
>>>
>>> The layering is not quite right here. NCP was essentially a
>>> transport protocol, and the IMPs provided lower-level network
>>> protocol services; in this sense, NCP is closer to TCP than to
>>> IP. The IMPs, in turn, used a protocol that was commonly called
>>> "1822" (from a BBN technical report) to communicate with ARPANET
>>> hosts; the initial TCP/IP implementations hosted on ARPANET fed
>>> IP datagrams directly to IMPs using 1822.
>>>
>>> See, e.g., IEN 28, sec 1.4 ["Interfaces"]. To quote:
>>> |In the ARPANET case, for example, the Internet module would
>>> |call on a local net module which would add the 1822 leader [6]
>>> |to the internet segment creating an ARPANET message to transmit
>>> |to the IMP.
>>> (From: https://www.rfc-editor.org/ien/ien28.pdf)
>>>
>>> The ARPANET was the first backbone for internetworking using IP,
>>> but TCP/IP and NCP sort of existed in quasi-parallel at the
>>> time. So TCP/IP hosts were very much "on the ARPANET", in the
>>> sense that they used the packet network of IMPs for
>>> communication in the same way that NCP-only hosts did.
>>
>> It again goes into what do we mean when we say "ARPNANET".
>
> The historical record shows that the players at the time meant
> the network of IMPs and the hosts that connected to them. It
> seems pretty clear that they didn't _just_ mean NCP.

I would argue that ARPANET was the host and the services they provided.
Just become something else went over the same cables don't mean anything
meaningful.

>> Just because you had other protocols using the same underlying
>> infrastructure, does it mean they are part of the same network?
>> I would say not.
>
> This is arguing semantics to an extent, but to answer this
> question, I would describe such an arrangement as different
> applications of the underlying network.

I would disagree that it's semantics. If you took a computer that talked
TCP/IP and hooked it up to an IMP before flag day, you would be unable
to communcate with all the hosts on ARPANET, even if they were at the
other end of that IMP.

You could talk to other machines that talked TCP/IP, but to reach any
resources on what people referred to as ARPANET you would need a gateway
that translated your traffic, or content, to something that could go
over NCP. If there was no gateway, you were essentially isolated as your
own host, no matter how much of ARPANET was carried over the same IMP.

Johnny

On 2023-10-03 07:55, gah4 wrote:
> On Monday, October 2, 2023 at 10:43:19 PM UTC-7, jimc...@gmail.com wrote:
>
> (snip)
>
>> The issue with VAX 9000 wasn't that it was an implementation of the VAX CISC architecture; the issue was chasing IBM with a massive ECL implementation of VAX, with all the associated costs in power and cooling and engineering required to address them. By the time VAX 9000 launched, CMOS VAX was already faster and dramatically cheaper.

More or less agree. It's not that the NVAX (the last CMOS VAX) was
faster. But it was close to the same speed as the 9000 at a fraction of
the cost, power requirements and size. And it was the later improved
giving the NVAX+ and NVAX++ which were faster.

That in combination with the 9000 taking way longer than planned to get
to market meant that there was no business case for the 9000 when it
came out.

> In 1978, CMOS was slower than TTL, and harder to build than NMOS.
>
> So, there is a transition when CMOS gets faster, and lower power,
> (and power density) the transition was made.

I think the writing was already on the wall while the 9000 was being
developed. So it was just a question of time, and with the 9000 being
late, this became even more an issue.

> The other problem, at least for some years, with CMOS is
> parasitic SCRs. The way the PN and NP junctions combine,
> can lead to the configuration of an SCR. If you manage to
> turn it in, it is a direct connection across the power supply,
> sometimes destructively.
>
> I don't remember the timeline for CMOS VAX, though.
> (Even though I have a MicroVAX 2000 and 3100.)

I think all single chip VAXen were CMOS. Not at all sure about the uVAX
I, but the II was, I think. Later ones definitely. Which are all before
the 9000.

The 9000 came out about the same time as the NVAX, which was the last
new VAX design in CMOS.

Johnny

In article <ufh5rl$fi6$2@news.misty.com>,
Johnny Billquist <bqt@softjar.se> wrote:
>On 2023-10-02 17:40, Dan Cross wrote:
>>[snip]
>> The historical record shows that the players at the time meant
>> the network of IMPs and the hosts that connected to them. It
>> seems pretty clear that they didn't _just_ mean NCP.
>
>I would argue that ARPANET was the host and the services they provided.
>Just become something else went over the same cables don't mean anything
>meaningful.

I mean, you have the words of the people involved with respect
to what they meant. They clearly referred to IP going over "the
ARPANET" in IEN 28, among other contemporary accounts. We can
sit here, 40 years after the fact, and spitball about what they
_really_ meant or how they were wrong all we want, but we can
see directly what they were referring to.

>>> Just because you had other protocols using the same underlying
>>> infrastructure, does it mean they are part of the same network?
>>> I would say not.
>>
>> This is arguing semantics to an extent, but to answer this
>> question, I would describe such an arrangement as different
>> applications of the underlying network.
>
>I would disagree that it's semantics. If you took a computer that talked
>TCP/IP and hooked it up to an IMP before flag day, you would be unable
>to communcate with all the hosts on ARPANET, even if they were at the
>other end of that IMP.
>
>You could talk to other machines that talked TCP/IP, but to reach any
>resources on what people referred to as ARPANET you would need a gateway
>that translated your traffic, or content, to something that could go
>over NCP. If there was no gateway, you were essentially isolated as your
>own host, no matter how much of ARPANET was carried over the same IMP.

We have IPv4-only hosts on the Internet today that cannot
communicate with IPv6 hosts unless through a gateway of some
kind; would you argue that IPv6-only hosts are therefore not
"on the Internet"?

There were machines on the ARPANET before NCP was invented;
presumably some didn't even speak NCP after it was invented.
Were the first machines on the ARPANET therefore not on the
ARPANET because they didn't speak NCP?

- Dan C.

On 2023-10-03 16:11, Dan Cross wrote:
> In article <ufh5rl$fi6$2@news.misty.com>,
> Johnny Billquist <bqt@softjar.se> wrote:
>> On 2023-10-02 17:40, Dan Cross wrote:
>>> [snip]
>>> The historical record shows that the players at the time meant
>>> the network of IMPs and the hosts that connected to them. It
>>> seems pretty clear that they didn't _just_ mean NCP.
>>
>> I would argue that ARPANET was the host and the services they provided.
>> Just become something else went over the same cables don't mean anything
>> meaningful.
>
> I mean, you have the words of the people involved with respect
> to what they meant. They clearly referred to IP going over "the
> ARPANET" in IEN 28, among other contemporary accounts. We can
> sit here, 40 years after the fact, and spitball about what they
> _really_ meant or how they were wrong all we want, but we can
> see directly what they were referring to.

That document talks about a theoretical ARPANET running TCP. Which you
could argue is what happened after flag day.

And the addressing scheme/ideas in that document is also an interesting
read. It's obviously different than what eventually was defined in IP.

So is this document relevant to bring up here? It's not something that
ever actually existed, but was the start of the process that eventually
led to the switch at flag day to TCP/IP.

>>>> Just because you had other protocols using the same underlying
>>>> infrastructure, does it mean they are part of the same network?
>>>> I would say not.
>>>
>>> This is arguing semantics to an extent, but to answer this
>>> question, I would describe such an arrangement as different
>>> applications of the underlying network.
>>
>> I would disagree that it's semantics. If you took a computer that talked
>> TCP/IP and hooked it up to an IMP before flag day, you would be unable
>> to communcate with all the hosts on ARPANET, even if they were at the
>> other end of that IMP.
>>
>> You could talk to other machines that talked TCP/IP, but to reach any
>> resources on what people referred to as ARPANET you would need a gateway
>> that translated your traffic, or content, to something that could go
>> over NCP. If there was no gateway, you were essentially isolated as your
>> own host, no matter how much of ARPANET was carried over the same IMP.
>
> We have IPv4-only hosts on the Internet today that cannot
> communicate with IPv6 hosts unless through a gateway of some
> kind; would you argue that IPv6-only hosts are therefore not
> "on the Internet"?

Well. At the moment, IPv6 only hosts don't really exist yet, but the
time might (will?) come. Eventually, I expect IPv4 to be phased out, at
which point an IPv4 only host will not be on the Inetnet anymore.
But in a sense yes, we're sort of getting to a dual-protocol Internet at
the moment. Fallback for most anyone/anything is still IPv4.

> There were machines on the ARPANET before NCP was invented;
> presumably some didn't even speak NCP after it was invented.
> Were the first machines on the ARPANET therefore not on the
> ARPANET because they didn't speak NCP?

If ARPANET was talking some other protocol before NCP, then obviously
that was the protocol you needed to talk to be on ARPANET, not NCP.
(I honestly don't know if there was something before NCP.)

Johnny

In article <ufh8jd$fi6$5@news.misty.com>,
Johnny Billquist <bqt@softjar.se> wrote:
>On 2023-10-03 16:11, Dan Cross wrote:
>> In article <ufh5rl$fi6$2@news.misty.com>,
>> Johnny Billquist <bqt@softjar.se> wrote:
>>> On 2023-10-02 17:40, Dan Cross wrote:
>>>> [snip]
>>>> The historical record shows that the players at the time meant
>>>> the network of IMPs and the hosts that connected to them. It
>>>> seems pretty clear that they didn't _just_ mean NCP.
>>>
>>> I would argue that ARPANET was the host and the services they provided.
>>> Just become something else went over the same cables don't mean anything
>>> meaningful.
>>
>> I mean, you have the words of the people involved with respect
>> to what they meant. They clearly referred to IP going over "the
>> ARPANET" in IEN 28, among other contemporary accounts. We can
>> sit here, 40 years after the fact, and spitball about what they
>> _really_ meant or how they were wrong all we want, but we can
>> see directly what they were referring to.
>
>That document talks about a theoretical ARPANET running TCP. Which you
>could argue is what happened after flag day.

No...It talks about sending "Internet Protocol" "segments" over
the ARPANET using the 1822 protocol. It says it right there on
the tin.

Note that this is not just TCP; this is actually IP. IEN 2
suggested layering into IP and TCP:
https://www.rfc-editor.org/ien/ien2.txt

>And the addressing scheme/ideas in that document is also an interesting
>read. It's obviously different than what eventually was defined in IP.

Well, yes: this was IPv2, which was an experimental version. So
what?

>So is this document relevant to bring up here? It's not something that
>ever actually existed, but was the start of the process that eventually
>led to the switch at flag day to TCP/IP.

Well, the part that I quoted talked about sending IP datagrams
over the ARPANET by wrapping the in 1822 frames and sending them
to an IMP. I'd say that's relevant with respect to exploring
what the authors of the early IP drafts were thinking: they had
a network, that network (which the called the "ARPANET") could
talk NCP, but they also obviously felt that they could make it
talk IP/TCP as well.

>>>>> Just because you had other protocols using the same underlying
>>>>> infrastructure, does it mean they are part of the same network?
>>>>> I would say not.
>>>>
>>>> This is arguing semantics to an extent, but to answer this
>>>> question, I would describe such an arrangement as different
>>>> applications of the underlying network.
>>>
>>> I would disagree that it's semantics. If you took a computer that talked
>>> TCP/IP and hooked it up to an IMP before flag day, you would be unable
>>> to communcate with all the hosts on ARPANET, even if they were at the
>>> other end of that IMP.
>>>
>>> You could talk to other machines that talked TCP/IP, but to reach any
>>> resources on what people referred to as ARPANET you would need a gateway
>>> that translated your traffic, or content, to something that could go
>>> over NCP. If there was no gateway, you were essentially isolated as your
>>> own host, no matter how much of ARPANET was carried over the same IMP.
>>
>> We have IPv4-only hosts on the Internet today that cannot
>> communicate with IPv6 hosts unless through a gateway of some
>> kind; would you argue that IPv6-only hosts are therefore not
>> "on the Internet"?
>
>Well. At the moment, IPv6 only hosts don't really exist yet, but the
>time might (will?) come.

Um, sure they do. Plenty of IoT and embedded devices have
skipped v4 entirely.

>Eventually, I expect IPv4 to be phased out, at
>which point an IPv4 only host will not be on the Inetnet anymore.
>But in a sense yes, we're sort of getting to a dual-protocol Internet at
>the moment. Fallback for most anyone/anything is still IPv4.

Ah, but both are called the "Internet"? Noted. :-)

>> There were machines on the ARPANET before NCP was invented;
>> presumably some didn't even speak NCP after it was invented.
>> Were the first machines on the ARPANET therefore not on the
>> ARPANET because they didn't speak NCP?
>
>If ARPANET was talking some other protocol before NCP, then obviously
>that was the protocol you needed to talk to be on ARPANET, not NCP.
>(I honestly don't know if there was something before NCP.)

This is easily discoverable. In addition to my note earlier in
this thread about the Host<->protocol known as "1822"
(https://groups.google.com/g/comp.os.vms/c/aX_f3g9O9jo/m/HMYxbVsRAgAJ),
one can simply look at the relevant RFCs: RFC 33 describes NCP:
https://datatracker.ietf.org/doc/html/rfc33

RFC11 describes the earlier host-host protocol:
https://datatracker.ietf.org/doc/html/rfc11
(which in turn refers to BBN report 1822)

Anyway, it seems clear from the historical record that the
people working on TCP/IP thought of the ARPANET as somehow
distinct from just hosts using NCP. You may chose to disagree,
but I don't see any evidence that that's how any of the players
at the time thought of it, and indeed, I see evidence to the
contrary.

- Dan C.

On 10/3/2023 9:50 AM, Johnny Billquist wrote:
> On 2023-10-03 07:55, gah4 wrote:
>> On Monday, October 2, 2023 at 10:43:19 PM UTC-7, jimc...@gmail.com wrote:
>>> The issue with VAX 9000 wasn't that it was an implementation of the
>>> VAX CISC architecture; the issue was chasing IBM with a massive ECL
>>> implementation of VAX, with all the associated costs in power and
>>> cooling and engineering required to address them. By the time VAX
>>> 9000 launched, CMOS VAX was already faster and dramatically cheaper.
>
> More or less agree. It's not that the NVAX (the last CMOS VAX) was
> faster. But it was close to the same speed as the 9000 at a fraction of
> the cost, power requirements and size. And it was the later improved
> giving the NVAX+ and NVAX++ which were faster.
>
> That in combination with the 9000 taking way longer than planned to get
> to market meant that there was no business case for the 9000 when it
> came out.
>
>> In 1978, CMOS was slower than TTL, and harder to build than NMOS.
>>
>> So, there is a transition when CMOS gets faster, and lower power,
>> (and power density) the transition was made.
>
> I think the writing was already on the wall while the 9000 was being
> developed. So it was just a question of time, and with the 9000 being
> late, this became even more an issue.
>
>> The other problem, at least for some years, with CMOS is
>> parasitic SCRs. The way the PN and NP junctions combine,
>> can lead to the configuration of an SCR.  If you manage to
>> turn it in, it is a direct connection across the power supply,
>> sometimes destructively.
>>
>> I don't remember the timeline for CMOS VAX, though.
>> (Even though I have a MicroVAX 2000 and 3100.)
>
> I think all single chip VAXen were CMOS. Not at all sure about the uVAX
> I, but the II was, I think. Later ones definitely. Which are all before
> the 9000.
>
> The 9000 came out about the same time as the NVAX, which was the last
> new VAX design in CMOS.

It is a fact that "price per VUPS" was very high for the 9000
compared to smaller VAX'es.

But was it intended to compete on "price per VUPS"?

I would have thought that it was intended to compete on:
* max CPU in a single box
* max RAM in a single box
* max IO capacity in a single box

And just maybe its fast demise was also due to the fact that the
mainframe market was moving to a single architecture (IBM mainframe
with IBM, Amdahl and Hitachi as vendors).

Arne

On 2023-10-03 17:45, Dan Cross wrote:
> In article <ufh8jd$fi6$5@news.misty.com>,
> Johnny Billquist <bqt@softjar.se> wrote:
>> On 2023-10-03 16:11, Dan Cross wrote:
>>> In article <ufh5rl$fi6$2@news.misty.com>,
>>> Johnny Billquist <bqt@softjar.se> wrote:
>>>> On 2023-10-02 17:40, Dan Cross wrote:
>>>>> [snip]
>>>>> The historical record shows that the players at the time meant
>>>>> the network of IMPs and the hosts that connected to them. It
>>>>> seems pretty clear that they didn't _just_ mean NCP.
>>>>
>>>> I would argue that ARPANET was the host and the services they provided.
>>>> Just become something else went over the same cables don't mean anything
>>>> meaningful.
>>>
>>> I mean, you have the words of the people involved with respect
>>> to what they meant. They clearly referred to IP going over "the
>>> ARPANET" in IEN 28, among other contemporary accounts. We can
>>> sit here, 40 years after the fact, and spitball about what they
>>> _really_ meant or how they were wrong all we want, but we can
>>> see directly what they were referring to.
>>
>> That document talks about a theoretical ARPANET running TCP. Which you
>> could argue is what happened after flag day.
>
> No...It talks about sending "Internet Protocol" "segments" over
> the ARPANET using the 1822 protocol. It says it right there on
> the tin.
>
> Note that this is not just TCP; this is actually IP. IEN 2
> suggested layering into IP and TCP:
> https://www.rfc-editor.org/ien/ien2.txt

Yes. And to quote that document:

An analogy may be drawn between the internet situation and the
ARPANET. The endpoints of message transmissions are hosts in both
cases, and they exchange messages conforming to a host to host
protocol. In the ARPA subnet there is a IMP to IMP protocol that is
primarily a hop by hop protocol, to parallel this the internet system
should have a hop by hop internet protocol. In the ARPANET a host and
an IMP interact through an inteface, commonly called 1822, which
specifies the format of messages crossing the boundary, an equivalent
interface in needed in the internet system.

Internet != ARPANET. IMPs are dealing with hop by hop communication.
Host protocol (NCP in this case) is dealing with host to host
communication, which in the internet case is TCP in development.

Anyway - I think we've beaten this horse to death, and I have a feeling
neither of us will convince the other. And that means further
discussions will only be more noise for others.

Feel free to comment and get last words in. I'll try to stop from my side.

Johnny

On 10/3/2023 7:22 AM, Dan Cross wrote:
> In article <b767aa9a-e341-426e-8215-0a9e9f6697bcn@googlegroups.com>,
> jimc...@gmail.com <jimcausey@gmail.com> wrote:
>> On Monday, September 25, 2023 at 5:15:50 PM UTC-7, Dan Cross wrote:
>>> The early days of Windows NT are well-documented in the book,
>>> "Show Stopper!" by G. Pascal Zachary. In short, they used OS/2
>>> and 386 machines; NT was self-hosting within a couple of years.
>>
>> Zachary wasn't very technical and made a number of mistakes in that book, although overall his story is well-researched and compelling. NT
>> was originally brought up on single-board Intel i860 hardware, followed by MIPS DECstations and then i386 hardware; Cutler insisted that the
>> team not focus on i386 because he wanted to keep NT from becoming wedded to the x86 architecture.
>
> I knew about the i860, but this is the first I've heard about
> the DECstation port (I thought they used MIPS Magnums or
> something?).

Most sources talk about just the CPU: MIPS R3000.

Per:

https://en.wikipedia.org/wiki/DECstation
https://en.wikipedia.org/wiki/MIPS_Magnum
https://en.wikipedia.org/wiki/Jazz_(computer)
https://www.linux-mips.org/wiki/Jazz

then:
- MIPS had a Magnum R3000 system with R3000 CPU
and TurboChannel bus
- DEC has a DECstatiosn 5000 system with R3000 CPU
and TurboChannel bus
- MS developed their own Jazz system with R3000 CPU
and EISA bus for use by Windows NT
- MS sold Jazz to MIPS that turned it into
Magnum R4000 with R4000 CPU and EISA bus (and was
big endian unlike MS Jazz that was little endian)

Arne

On 2023-10-03 22:18, Arne Vajhøj wrote:
> On 10/3/2023 9:50 AM, Johnny Billquist wrote:
>> On 2023-10-03 07:55, gah4 wrote:
>>> On Monday, October 2, 2023 at 10:43:19 PM UTC-7, jimc...@gmail.com
>>> wrote:
>>>> The issue with VAX 9000 wasn't that it was an implementation of the
>>>> VAX CISC architecture; the issue was chasing IBM with a massive ECL
>>>> implementation of VAX, with all the associated costs in power and
>>>> cooling and engineering required to address them. By the time VAX
>>>> 9000 launched, CMOS VAX was already faster and dramatically cheaper.
>>
>> More or less agree. It's not that the NVAX (the last CMOS VAX) was
>> faster. But it was close to the same speed as the 9000 at a fraction
>> of the cost, power requirements and size. And it was the later
>> improved giving the NVAX+ and NVAX++ which were faster.
>>
>> That in combination with the 9000 taking way longer than planned to
>> get to market meant that there was no business case for the 9000 when
>> it came out.
>>
>>> In 1978, CMOS was slower than TTL, and harder to build than NMOS.
>>>
>>> So, there is a transition when CMOS gets faster, and lower power,
>>> (and power density) the transition was made.
>>
>> I think the writing was already on the wall while the 9000 was being
>> developed. So it was just a question of time, and with the 9000 being
>> late, this became even more an issue.
>>
>>> The other problem, at least for some years, with CMOS is
>>> parasitic SCRs. The way the PN and NP junctions combine,
>>> can lead to the configuration of an SCR.  If you manage to
>>> turn it in, it is a direct connection across the power supply,
>>> sometimes destructively.
>>>
>>> I don't remember the timeline for CMOS VAX, though.
>>> (Even though I have a MicroVAX 2000 and 3100.)
>>
>> I think all single chip VAXen were CMOS. Not at all sure about the
>> uVAX I, but the II was, I think. Later ones definitely. Which are all
>> before the 9000.
>>
>> The 9000 came out about the same time as the NVAX, which was the last
>> new VAX design in CMOS.
>
> It is a fact that "price per VUPS" was very high for the 9000
> compared to smaller VAX'es.

Yes.

> But was it intended to compete on "price per VUPS"?

Well. The problem was that when the 9000 finally did come out, it was
not competitive from any perspective.
It was way more expensive than a 7000. It was way larger than a 7000. It
was way costlier to run than a 7000. It had close to similar performance
to a 7000. The 9000 started shipping in 1991, while the 7000 shipped in
1992.
Why would anyone buy the 9000? There was just a small window left before
the NVAX based 7000 came. And everyone knew that was coming. NVAX based
machines as such started shipping in 1991 as well.

> I would have thought that it was intended to compete on:
> * max CPU in a single box

NVAX does it better.

> * max RAM in a single box

NVAX does it better.

> * max IO capacity in a single box

In a single box, I think they come out even. The 9000 had massively more
I/O capacity, if you look at the full system. But that's a lot of boxes.

And DEC was also pushing for clusters, and had been for quite a while,
where the capacity of a single machine wasn't the main point.

> And just maybe its fast demise was also due to the fact that the
> mainframe market was moving to a single architecture (IBM mainframe
> with IBM, Amdahl and Hitachi as vendors).

Possibly, but I wouldn't think so. There were plenty of DEC customers
looking for faster VAXen. And the VAX market was still strong at that
time, athough the Alpha was just about coming in as well.

Johnny

Den 2023-10-04 kl. 12:40, skrev Johnny Billquist:
> On 2023-10-03 22:18, Arne Vajhøj wrote:
>> On 10/3/2023 9:50 AM, Johnny Billquist wrote:
>>> On 2023-10-03 07:55, gah4 wrote:
>>>> On Monday, October 2, 2023 at 10:43:19 PM UTC-7, jimc...@gmail.com wrote:
>>>>> The issue with VAX 9000 wasn't that it was an implementation of the
>>>>> VAX CISC architecture; the issue was chasing IBM with a massive ECL
>>>>> implementation of VAX, with all the associated costs in power and
>>>>> cooling and engineering required to address them. By the time VAX 9000
>>>>> launched, CMOS VAX was already faster and dramatically cheaper.
>>>
>>> More or less agree. It's not that the NVAX (the last CMOS VAX) was
>>> faster. But it was close to the same speed as the 9000 at a fraction of
>>> the cost, power requirements and size. And it was the later improved
>>> giving the NVAX+ and NVAX++ which were faster.
>>>
>>> That in combination with the 9000 taking way longer than planned to get
>>> to market meant that there was no business case for the 9000 when it
>>> came out.
>>>
>>>> In 1978, CMOS was slower than TTL, and harder to build than NMOS.
>>>>
>>>> So, there is a transition when CMOS gets faster, and lower power,
>>>> (and power density) the transition was made.
>>>
>>> I think the writing was already on the wall while the 9000 was being
>>> developed. So it was just a question of time, and with the 9000 being
>>> late, this became even more an issue.
>>>
>>>> The other problem, at least for some years, with CMOS is
>>>> parasitic SCRs. The way the PN and NP junctions combine,
>>>> can lead to the configuration of an SCR.  If you manage to
>>>> turn it in, it is a direct connection across the power supply,
>>>> sometimes destructively.
>>>>
>>>> I don't remember the timeline for CMOS VAX, though.
>>>> (Even though I have a MicroVAX 2000 and 3100.)
>>>
>>> I think all single chip VAXen were CMOS. Not at all sure about the uVAX
>>> I, but the II was, I think. Later ones definitely. Which are all before
>>> the 9000.
>>>
>>> The 9000 came out about the same time as the NVAX, which was the last
>>> new VAX design in CMOS.
>>
>> It is a fact that "price per VUPS" was very high for the 9000
>> compared to smaller VAX'es.
>
> Yes.
>
>> But was it intended to compete on "price per VUPS"?
>
> Well. The problem was that when the 9000 finally did come out, it was not
> competitive from any perspective.
> It was way more expensive than a 7000. It was way larger than a 7000. It
> was way costlier to run than a 7000. It had close to similar performance to
> a 7000. The 9000 started shipping in 1991, while the 7000 shipped in 1992.
> Why would anyone buy the 9000? There was just a small window left before
> the NVAX based 7000 came. And everyone knew that was coming. NVAX based
> machines as such started shipping in 1991 as well.
>
>> I would have thought that it was intended to compete on:
>> * max CPU in a single box
>
> NVAX does it better.
>
>> * max RAM in a single box
>
> NVAX does it better.
>
>> * max IO capacity in a single box
>
> In a single box, I think they come out even. The 9000 had massively more
> I/O capacity, if you look at the full system. But that's a lot of boxes.
>
> And DEC was also pushing for clusters, and had been for quite a while,
> where the capacity of a single machine wasn't the main point.
>
>> And just maybe its fast demise was also due to the fact that the
>> mainframe market was moving to a single architecture (IBM mainframe
>> with IBM, Amdahl and Hitachi as vendors).
>
> Possibly, but I wouldn't think so. There were plenty of DEC customers
> looking for faster VAXen. And the VAX market was still strong at that time,
> athough the Alpha was just about coming in as well.
>
>   Johnny
>

If I'm not completaly wrong, the Swedish weather service (SMHI) had one
VAX 9000 once. I seam to remember seeing it behind bars and fences at a
visit to the SMHI HQ in Norrköping/Sweden. No one but DEC people was
allowed to touch it. This was around the same time as a scandal with
a 11/782 (I think) that was going to ship to Soviet.

I also found this note from 1994:
"1994-01-01, Over the last four years, CERN has progressively converted its
central batch production facilities from classic mainframe platforms (Cray
XMP, IBM, ESA, Vax 9000) to distributed RISC based facilities..."

On 2023-10-04 14:49, Jan-Erik Söderholm wrote:
> If I'm not completaly wrong, the Swedish weather service (SMHI) had one
> VAX 9000 once. I seam to remember seeing it behind bars and fences at a
> visit to the SMHI HQ in Norrköping/Sweden. No one but DEC people was
> allowed to touch it. This was around the same time as a scandal with
> a 11/782 (I think) that was going to ship to Soviet.

I'm old enough to remember the VAX-11/782 scandal in fairly vivid
detail. And it was long before the 9000 was even thought of. (A good
summary exists on the Swedish wikipedia page:
https://sv.wikipedia.org/wiki/Containeraff%C3%A4ren. Google can probably
do a good translation if anyone wants the details. But this was back in
1983.)

But I could possibly see that SMHI could have had one. They had big
computing requirements. I do know that SAAB in Linköping had one. I know
a guy who worked there at the time and was atleast somewhat responsible.
But the machine was gone before I got to know him. But he had kept some
memorabilia, and so I actually have a module from a VAX 9000 at home,
which came from the one at SAAB.

> I also found this note from 1994:
> "1994-01-01, Over the last four years, CERN has progressively converted
> its central batch production facilities from classic mainframe platforms
> (Cray XMP, IBM, ESA, Vax 9000) to distributed RISC based facilities..."

Interesting. Well, I guess it sortof makes sense that CERN would also be
a user, since they also had extreme computation needs, and and money was
less of an issue.

Johnny

Den 2023-10-04 kl. 16:49, skrev Johnny Billquist:
> On 2023-10-04 14:49, Jan-Erik Söderholm wrote:
>> If I'm not completaly wrong, the Swedish weather service (SMHI) had one
>> VAX 9000 once. I seam to remember seeing it behind bars and fences at a
>> visit to the SMHI HQ in Norrköping/Sweden. No one but DEC people was
>> allowed to touch it. This was around the same time as a scandal with
>> a 11/782 (I think) that was going to ship to Soviet.
>
> I'm old enough to remember the VAX-11/782 scandal in fairly vivid detail.
> And it was long before the 9000 was even thought of. (A good summary exists
> on the Swedish wikipedia page:
> https://sv.wikipedia.org/wiki/Containeraff%C3%A4ren. Google can probably do
> a good translation if anyone wants the details. But this was back in 1983.)
>
> But I could possibly see that SMHI could have had one. They had big
> computing requirements. I do know that SAAB in Linköping had one. I know a
> guy who worked there at the time and was atleast somewhat responsible. But
> the machine was gone before I got to know him. But he had kept some
> memorabilia, and so I actually have a module from a VAX 9000 at home, which
> came from the one at SAAB.
>
>   Johnny
>

As was described at the SMHI visit, their VAX systems was used as a kind
of "front-ends" against the Cray-1 system(s) at the center in Reading/UK
for weather calculations.

On 2023-10-04 17:11, Jan-Erik Söderholm wrote:
> Den 2023-10-04 kl. 16:49, skrev Johnny Billquist:
>> On 2023-10-04 14:49, Jan-Erik Söderholm wrote:
>>> If I'm not completaly wrong, the Swedish weather service (SMHI) had one
>>> VAX 9000 once. I seam to remember seeing it behind bars and fences at a
>>> visit to the SMHI HQ in Norrköping/Sweden. No one but DEC people was
>>> allowed to touch it. This was around the same time as a scandal with
>>> a 11/782 (I think) that was going to ship to Soviet.
>>
>> I'm old enough to remember the VAX-11/782 scandal in fairly vivid
>> detail. And it was long before the 9000 was even thought of. (A good
>> summary exists on the Swedish wikipedia page:
>> https://sv.wikipedia.org/wiki/Containeraff%C3%A4ren. Google can
>> probably do a good translation if anyone wants the details. But this
>> was back in 1983.)
>>
>> But I could possibly see that SMHI could have had one. They had big
>> computing requirements. I do know that SAAB in Linköping had one. I
>> know a guy who worked there at the time and was atleast somewhat
>> responsible. But the machine was gone before I got to know him. But he
>> had kept some memorabilia, and so I actually have a module from a VAX
>> 9000 at home, which came from the one at SAAB.
>>
>>    Johnny
>>
>
> As was described at the SMHI visit, their VAX systems was used as a kind
> of "front-ends" against the Cray-1 system(s) at the center in Reading/UK
> for weather calculations.

That makes it sound unlikely that it would have been VAX 9000 systems.

I'm trying to remember about VAX as front-end to Cray-1. And my brain
keeps trying to say something like VAX-11/780, which would also be more
aligned in time with the VAX-11/782 scandal. So I think you just got the
9000 mixed in there by mistake, and the rest probably sums up pretty
well. It could even have been a VAX-8600, which was the top dog in 1984.

Johnny

On 10/4/2023 6:40 AM, Johnny Billquist wrote:
> On 2023-10-03 22:18, Arne Vajhøj wrote:
>> On 10/3/2023 9:50 AM, Johnny Billquist wrote:
>>> On 2023-10-03 07:55, gah4 wrote:
>>>> On Monday, October 2, 2023 at 10:43:19 PM UTC-7, jimc...@gmail.com
>>>> wrote:
>>>>> The issue with VAX 9000 wasn't that it was an implementation of the
>>>>> VAX CISC architecture; the issue was chasing IBM with a massive ECL
>>>>> implementation of VAX, with all the associated costs in power and
>>>>> cooling and engineering required to address them. By the time VAX
>>>>> 9000 launched, CMOS VAX was already faster and dramatically cheaper.
>>>
>>> More or less agree. It's not that the NVAX (the last CMOS VAX) was
>>> faster. But it was close to the same speed as the 9000 at a fraction
>>> of the cost, power requirements and size. And it was the later
>>> improved giving the NVAX+ and NVAX++ which were faster.
>>>
>>> That in combination with the 9000 taking way longer than planned to
>>> get to market meant that there was no business case for the 9000 when
>>> it came out.
>>>
>>>> In 1978, CMOS was slower than TTL, and harder to build than NMOS.
>>>>
>>>> So, there is a transition when CMOS gets faster, and lower power,
>>>> (and power density) the transition was made.
>>>
>>> I think the writing was already on the wall while the 9000 was being
>>> developed. So it was just a question of time, and with the 9000 being
>>> late, this became even more an issue.
>>>
>>>> The other problem, at least for some years, with CMOS is
>>>> parasitic SCRs. The way the PN and NP junctions combine,
>>>> can lead to the configuration of an SCR.  If you manage to
>>>> turn it in, it is a direct connection across the power supply,
>>>> sometimes destructively.
>>>>
>>>> I don't remember the timeline for CMOS VAX, though.
>>>> (Even though I have a MicroVAX 2000 and 3100.)
>>>
>>> I think all single chip VAXen were CMOS. Not at all sure about the
>>> uVAX I, but the II was, I think. Later ones definitely. Which are all
>>> before the 9000.
>>>
>>> The 9000 came out about the same time as the NVAX, which was the last
>>> new VAX design in CMOS.
>>
>> It is a fact that "price per VUPS" was very high for the 9000
>> compared to smaller VAX'es.
>
> Yes.
>
>> But was it intended to compete on "price per VUPS"?
>
> Well. The problem was that when the 9000 finally did come out, it was
> not competitive from any perspective.
> It was way more expensive than a 7000. It was way larger than a 7000. It
> was way costlier to run than a 7000. It had close to similar performance
> to a 7000. The 9000 started shipping in 1991, while the 7000 shipped in
> 1992.

I think 9000 started shipping in 1990.

> Why would anyone buy the 9000? There was just a small window left before
> the NVAX based 7000 came. And everyone knew that was coming. NVAX based
> machines as such started shipping in 1991 as well.
>
>> I would have thought that it was intended to compete on:
>> * max CPU in a single box
>
> NVAX does it better.
>
>> * max RAM in a single box
>
> NVAX does it better.
>
>> * max IO capacity in a single box
>
> In a single box, I think they come out even. The 9000 had massively more
> I/O capacity, if you look at the full system. But that's a lot of boxes.

As I recall it then the 6000 was a single cabinet wide thing, while
a 9000 (at least in large config - 400??) was a 3 cabinet wide thing.

> And DEC was also pushing for clusters, and had been for quite a while,
> where the capacity of a single machine wasn't the main point.

In the VAX market.

The mainframe market was still single machine oriented.

>> And just maybe its fast demise was also due to the fact that the
>> mainframe market was moving to a single architecture (IBM mainframe
>> with IBM, Amdahl and Hitachi as vendors).
>
> Possibly, but I wouldn't think so. There were plenty of DEC customers
> looking for faster VAXen. And the VAX market was still strong at that
> time, athough the Alpha was just about coming in as well.

I am not sure that we really disagree so much.

The 9000 did not have a market.

In the "super-super mini-computer" market it was too expensive.
You could buy a handful of 6000's or huge number of MicroVAX'es
for the price of a 9000. A 9000 was simply not cost efficient
in that market.

In the "mainframe" market (it was branded as an IBM mainframe
killer so that market must have been considered relevant) the
time of the non-IBM-compatible mainframe was over.

And in the "super computer" market then one could buy the vector
bolton for the 9000 (and I suspect some did - the swedish weather
service 9000 Jan-Erik remembers probably had it), but it was still
a very expensive system - and the 6000 could also get vector
bolton.

Arne

Johnny Billquist <bqt@softjar.se> wrote:
>On 2023-10-03 01:58, Scott Dorsey wrote:
>> gah4 <gah4@u.washington.edu> wrote:
>>> It is some years now, so I don't remember the details, but I am pretty
>>> sure that there was one that worked even if you didn't have an account.
>>
>> Decnet to arpa? Sure, there were lots of them and none that I know
>> required an account. It was just a polite service people provided.
>> The best one was at Columbia which had really good connectivity (and also
>> bitnet connectivity) so you could do "fredbox::fred@columbia.edu" as I
>> recall.
>
>For mail, yes.
>
>That still happens... Try sending to "pondus::bqt"@mim.stupi.net and
>you'll reach me on my PDP-11/93 running RSX-11M-PLUS at home.

Excellent! Do you gateway to bitnet too?

For remote login, there were lots of computers out there that ran both
tcp/ip and decnet... these included lots of vaxen running vms and ultrix
but also included Sun systems and others. Get an account on any machine
with both of them (or use the free guest account at ai.mit.edu) and you
can telnet into one machine and set host out of it to another.
--scott

--
"C'est un Nagra. C'est suisse, et tres, tres precis."

On 2023-10-04 23:00, Scott Dorsey wrote:
> Johnny Billquist <bqt@softjar.se> wrote:
>> On 2023-10-03 01:58, Scott Dorsey wrote:
>>> gah4 <gah4@u.washington.edu> wrote:
>>>> It is some years now, so I don't remember the details, but I am pretty
>>>> sure that there was one that worked even if you didn't have an account.
>>>
>>> Decnet to arpa? Sure, there were lots of them and none that I know
>>> required an account. It was just a polite service people provided.
>>> The best one was at Columbia which had really good connectivity (and also
>>> bitnet connectivity) so you could do "fredbox::fred@columbia.edu" as I
>>> recall.
>>
>> For mail, yes.
>>
>> That still happens... Try sending to "pondus::bqt"@mim.stupi.net and
>> you'll reach me on my PDP-11/93 running RSX-11M-PLUS at home.
>
> Excellent! Do you gateway to bitnet too?

Nope. Never did bitnet, and have basically zero knowledge about it,
apart from knowing it existed.

> For remote login, there were lots of computers out there that ran both
> tcp/ip and decnet... these included lots of vaxen running vms and ultrix
> but also included Sun systems and others. Get an account on any machine
> with both of them (or use the free guest account at ai.mit.edu) and you
> can telnet into one machine and set host out of it to another.

Yeah. That's what you normally would have to do. No direct translation
between the protocols.

Johnny

On 2023-10-04 22:59, Arne Vajhøj wrote:
> On 10/4/2023 6:40 AM, Johnny Billquist wrote:
>> On 2023-10-03 22:18, Arne Vajhøj wrote:
>>> On 10/3/2023 9:50 AM, Johnny Billquist wrote:
>>>> The 9000 came out about the same time as the NVAX, which was the
>>>> last new VAX design in CMOS.
>>>
>>> It is a fact that "price per VUPS" was very high for the 9000
>>> compared to smaller VAX'es.
>>
>> Yes.
>>
>>> But was it intended to compete on "price per VUPS"?
>>
>> Well. The problem was that when the 9000 finally did come out, it was
>> not competitive from any perspective.
>> It was way more expensive than a 7000. It was way larger than a 7000.
>> It was way costlier to run than a 7000. It had close to similar
>> performance to a 7000. The 9000 started shipping in 1991, while the
>> 7000 shipped in 1992.
>
> I think 9000 started shipping in 1990.

That was the original plan (or even 89), but they got delayed. Which was
part of the problem with the 9000. It was way expensive, and had serious
problems, and got delayed. I think initial deliveries were close to end
of 91, and it still had issues. According to Wikipedia a few systems
were shipped in 90, but they had issues.
(https://en.wikipedia.org/wiki/VAX_9000)

It was perhaps a hard sell even in the original plan, but with the
delays added, any window of opportunity was basically lost. The market
was definitely not there anymore when the systems finally were shipped.

>>> * max IO capacity in a single box
>>
>> In a single box, I think they come out even. The 9000 had massively
>> more I/O capacity, if you look at the full system. But that's a lot of
>> boxes.
>
> As I recall it then the 6000 was a single cabinet wide thing, while
> a 9000 (at least in large config - 400??) was a 3 cabinet wide thing.

The 9000 was, as far as I can recall, two double cabs and one single.
Maybe it was possible to get some smaller configs, but that compared to
the 6000 or 7000 at a single cab is quite a difference.

>> And DEC was also pushing for clusters, and had been for quite a while,
>> where the capacity of a single machine wasn't the main point.
>
> In the VAX market.
>
> The mainframe market was still single machine oriented.

True.

>>> And just maybe its fast demise was also due to the fact that the
>>> mainframe market was moving to a single architecture (IBM mainframe
>>> with IBM, Amdahl and Hitachi as vendors).
>>
>> Possibly, but I wouldn't think so. There were plenty of DEC customers
>> looking for faster VAXen. And the VAX market was still strong at that
>> time, athough the Alpha was just about coming in as well.
>
> I am not sure that we really disagree so much.

I'm not sure we disagree either. Mostly getting through the finer points
of the whole thing.

> The 9000 did not have a market.
>
> In the "super-super mini-computer" market it was too expensive.
> You could buy a handful of 6000's or huge number of MicroVAX'es
> for the price of a 9000. A 9000 was simply not cost efficient
> in that market.

Yes.

> In the "mainframe" market (it was branded as an IBM mainframe
> killer so that market must have been considered relevant) the
> time of the non-IBM-compatible mainframe was over.

Probably. But even so, the 9000 just wasn't the right choice for DEC.
But I guess in a way it was symptomatic of the whole IBM-ification of
DEC that started in the late 80s.

> And in the "super computer" market then one could buy the vector
> bolton for the 9000 (and I suspect some did - the swedish weather
> service 9000 Jan-Erik remembers probably had it), but it was still
> a very expensive system - and the 6000 could also get vector
> bolton.

Yup. But I don't think the vector option was ever very relevant. As far
as I can remember, it wasn't even available on the 6000-500 and
6000-600. Only the 6000-400. Which suggest that there wasn't any demand.
The 7000 never had it.

Johnny

On 10/4/2023 7:36 PM, Johnny Billquist wrote:
> On 2023-10-04 22:59, Arne Vajhøj wrote:
>> On 10/4/2023 6:40 AM, Johnny Billquist wrote:
>>> Well. The problem was that when the 9000 finally did come out, it was
>>> not competitive from any perspective.
>>> It was way more expensive than a 7000. It was way larger than a 7000.
>>> It was way costlier to run than a 7000. It had close to similar
>>> performance to a 7000. The 9000 started shipping in 1991, while the
>>> 7000 shipped in 1992.
>>
>> I think 9000 started shipping in 1990.
>
> That was the original plan (or even 89), but they got delayed. Which was
> part of the problem with the 9000. It was way expensive, and had serious
> problems, and got delayed. I think initial deliveries were close to end
> of 91, and it still had issues. According to Wikipedia a few systems
> were shipped in 90, but they had issues.
> (https://en.wikipedia.org/wiki/VAX_9000)

Usually shipping is considered shipping - problems or no problems.

>> In the "mainframe" market (it was branded as an IBM mainframe
>> killer so that market must have been considered relevant) the
>> time of the non-IBM-compatible mainframe was over.
>
> Probably. But even so, the 9000 just wasn't the right choice for DEC.
> But I guess in a way it was symptomatic of the whole IBM-ification of
> DEC that started in the late 80s.

When you are number two you want to be number one.

>> And in the "super computer" market then one could buy the vector
>> bolton for the 9000 (and I suspect some did - the swedish weather
>> service 9000 Jan-Erik remembers probably had it), but it was still
>> a very expensive system - and the 6000 could also get vector
>> bolton.
>
> Yup. But I don't think the vector option was ever very relevant. As far
> as I can remember, it wasn't even available on the 6000-500 and
> 6000-600. Only the 6000-400. Which suggest that there wasn't any demand.
> The 7000 never had it.

I think that is correct - only -400 and -500 had it.

VMS dropped out of the scientific computing market.

And soon after the entire vector thing went away in scientific
computing for a few decades until it came back in the form of GPU's.

Arne

On 10/4/2023 4:59 PM, Arne Vajhøj wrote:
> On 10/4/2023 6:40 AM, Johnny Billquist wrote:
>> On 2023-10-03 22:18, Arne Vajhøj wrote:
>>> On 10/3/2023 9:50 AM, Johnny Billquist wrote:
>>>> On 2023-10-03 07:55, gah4 wrote:
>>>>> On Monday, October 2, 2023 at 10:43:19 PM UTC-7, jimc...@gmail.com
>>>>> wrote:
>>>>>> The issue with VAX 9000 wasn't that it was an implementation of
>>>>>> the VAX CISC architecture; the issue was chasing IBM with a
>>>>>> massive ECL implementation of VAX, with all the associated costs
>>>>>> in power and cooling and engineering required to address them. By
>>>>>> the time VAX 9000 launched, CMOS VAX was already faster and
>>>>>> dramatically cheaper.
>>>>
>>>> More or less agree. It's not that the NVAX (the last CMOS VAX) was
>>>> faster. But it was close to the same speed as the 9000 at a fraction
>>>> of the cost, power requirements and size. And it was the later
>>>> improved giving the NVAX+ and NVAX++ which were faster.
>>>>
>>>> That in combination with the 9000 taking way longer than planned to
>>>> get to market meant that there was no business case for the 9000
>>>> when it came out.
>>>>
>>>>> In 1978, CMOS was slower than TTL, and harder to build than NMOS.
>>>>>
>>>>> So, there is a transition when CMOS gets faster, and lower power,
>>>>> (and power density) the transition was made.
>>>>
>>>> I think the writing was already on the wall while the 9000 was being
>>>> developed. So it was just a question of time, and with the 9000
>>>> being late, this became even more an issue.
>>>>
>>>>> The other problem, at least for some years, with CMOS is
>>>>> parasitic SCRs. The way the PN and NP junctions combine,
>>>>> can lead to the configuration of an SCR.  If you manage to
>>>>> turn it in, it is a direct connection across the power supply,
>>>>> sometimes destructively.
>>>>>
>>>>> I don't remember the timeline for CMOS VAX, though.
>>>>> (Even though I have a MicroVAX 2000 and 3100.)
>>>>
>>>> I think all single chip VAXen were CMOS. Not at all sure about the
>>>> uVAX I, but the II was, I think. Later ones definitely. Which are
>>>> all before the 9000.
>>>>
>>>> The 9000 came out about the same time as the NVAX, which was the
>>>> last new VAX design in CMOS.
>>>
>>> It is a fact that "price per VUPS" was very high for the 9000
>>> compared to smaller VAX'es.
>>
>> Yes.
>>
>>> But was it intended to compete on "price per VUPS"?
>>
>> Well. The problem was that when the 9000 finally did come out, it was
>> not competitive from any perspective.
>> It was way more expensive than a 7000. It was way larger than a 7000.
>> It was way costlier to run than a 7000. It had close to similar
>> performance to a 7000. The 9000 started shipping in 1991, while the
>> 7000 shipped in 1992.
>
> I think 9000 started shipping in 1990.
>
>> Why would anyone buy the 9000? There was just a small window left
>> before the NVAX based 7000 came. And everyone knew that was coming.
>> NVAX based machines as such started shipping in 1991 as well.
>>
>>> I would have thought that it was intended to compete on:
>>> * max CPU in a single box
>>
>> NVAX does it better.
>>
>>> * max RAM in a single box
>>
>> NVAX does it better.
>>
>>> * max IO capacity in a single box
>>
>> In a single box, I think they come out even. The 9000 had massively
>> more I/O capacity, if you look at the full system. But that's a lot of
>> boxes.
>
> As I recall it then the 6000 was a single cabinet wide thing, while
> a 9000 (at least in large config - 400??) was a 3 cabinet wide thing.
>
>> And DEC was also pushing for clusters, and had been for quite a while,
>> where the capacity of a single machine wasn't the main point.
>
> In the VAX market.
>
> The mainframe market was still single machine oriented.
>
>>> And just maybe its fast demise was also due to the fact that the
>>> mainframe market was moving to a single architecture (IBM mainframe
>>> with IBM, Amdahl and Hitachi as vendors).
>>
>> Possibly, but I wouldn't think so. There were plenty of DEC customers
>> looking for faster VAXen. And the VAX market was still strong at that
>> time, athough the Alpha was just about coming in as well.
>
> I am not sure that we really disagree so much.
>
> The 9000 did not have a market.
>
> In the "super-super mini-computer" market it was too expensive.
> You could buy a handful of 6000's or huge number of MicroVAX'es
> for the price of a 9000. A 9000 was simply not cost efficient
> in that market.
>
> In the "mainframe" market (it was branded as an IBM mainframe
> killer so that market must have been considered relevant) the
> time of the non-IBM-compatible mainframe was over.
>

I always find it funny when I see this comment.
Unisys (formerly UNIVAC) is doing just fine with their 2200 which
is the follow on from and compatible with the old 1100. As a
matter of fact, two of the largest ISes in use today run on
Unisys after running for more than a decade on the 1100.

bill

On 10/4/2023 8:11 PM, Arne Vajhøj wrote:
> On 10/4/2023 7:36 PM, Johnny Billquist wrote:
>> On 2023-10-04 22:59, Arne Vajhøj wrote:
>>> And in the "super computer" market then one could buy the vector
>>> bolton for the 9000 (and I suspect some did - the swedish weather
>>> service 9000 Jan-Erik remembers probably had it), but it was still
>>> a very expensive system - and the 6000 could also get vector
>>> bolton.
>>
>> Yup. But I don't think the vector option was ever very relevant. As
>> far as I can remember, it wasn't even available on the 6000-500 and
>> 6000-600. Only the 6000-400. Which suggest that there wasn't any demand.
>> The 7000 never had it.
>
> I think that is correct - only -400 and -500 had it.
>
> VMS dropped out of the scientific computing market.
>
> And soon after the entire vector thing went away in scientific
> computing for a few decades until it came back in the form of GPU's.

It probably did not help either that the vector bolton took
space away from CPU's.

Max was:
* 6 CPU + 0 vector
* 4 CPU + 1 vector
* 2 CPU + 2 vector

Arne

Ah the VAX-9000, nice machine if you could afford it, and you need that level of reliability.... it had some extra logical paths through all the 'logical gates' so the system could 'sense' if there was a problem that needed to be fixed. That was supposed to be a big selling point so they could work with FTVAX's for a high uptime solution.

Yup at the time the VAX-6000 was slower, but mot customer new about the faster VAX-6000's in the pipeline, and about the not yet released Alpha's.

Both the VAX-9000 and the FTVAX's main competitor was clustering... and even then we where looking at supporting terabyte disk farms and needed as much memory as was possible.

DEC had so many balls in the air, and no real clear strategy for software on the Alpha. Such a horrible mix of 32 and 64 bit components...

gt

On Wednesday, October 4, 2023 at 8:49:44 AM UTC-4, Jan-Erik Söderholm wrote:
> Den 2023-10-04 kl. 12:40, skrev Johnny Billquist:
> > On 2023-10-03 22:18, Arne Vajhøj wrote:
> >> On 10/3/2023 9:50 AM, Johnny Billquist wrote:
> >>> On 2023-10-03 07:55, gah4 wrote:
> >>>> On Monday, October 2, 2023 at 10:43:19 PM UTC-7, jimc...@gmail.com wrote:
> >>>>> The issue with VAX 9000 wasn't that it was an implementation of the
> >>>>> VAX CISC architecture; the issue was chasing IBM with a massive ECL
> >>>>> implementation of VAX, with all the associated costs in power and
> >>>>> cooling and engineering required to address them. By the time VAX 9000
> >>>>> launched, CMOS VAX was already faster and dramatically cheaper.
> >>>
> >>> More or less agree. It's not that the NVAX (the last CMOS VAX) was
> >>> faster. But it was close to the same speed as the 9000 at a fraction of
> >>> the cost, power requirements and size. And it was the later improved
> >>> giving the NVAX+ and NVAX++ which were faster.
> >>>
> >>> That in combination with the 9000 taking way longer than planned to get
> >>> to market meant that there was no business case for the 9000 when it
> >>> came out.
> >>>
> >>>> In 1978, CMOS was slower than TTL, and harder to build than NMOS.
> >>>>
> >>>> So, there is a transition when CMOS gets faster, and lower power,
> >>>> (and power density) the transition was made.
> >>>
> >>> I think the writing was already on the wall while the 9000 was being
> >>> developed. So it was just a question of time, and with the 9000 being
> >>> late, this became even more an issue.
> >>>
> >>>> The other problem, at least for some years, with CMOS is
> >>>> parasitic SCRs. The way the PN and NP junctions combine,
> >>>> can lead to the configuration of an SCR. If you manage to
> >>>> turn it in, it is a direct connection across the power supply,
> >>>> sometimes destructively.
> >>>>
> >>>> I don't remember the timeline for CMOS VAX, though.
> >>>> (Even though I have a MicroVAX 2000 and 3100.)
> >>>
> >>> I think all single chip VAXen were CMOS. Not at all sure about the uVAX
> >>> I, but the II was, I think. Later ones definitely. Which are all before
> >>> the 9000.
> >>>
> >>> The 9000 came out about the same time as the NVAX, which was the last
> >>> new VAX design in CMOS.
> >>
> >> It is a fact that "price per VUPS" was very high for the 9000
> >> compared to smaller VAX'es.
> >
> > Yes.
> >
> >> But was it intended to compete on "price per VUPS"?
> >
> > Well. The problem was that when the 9000 finally did come out, it was not
> > competitive from any perspective.
> > It was way more expensive than a 7000. It was way larger than a 7000. It
> > was way costlier to run than a 7000. It had close to similar performance to
> > a 7000. The 9000 started shipping in 1991, while the 7000 shipped in 1992.
> > Why would anyone buy the 9000? There was just a small window left before
> > the NVAX based 7000 came. And everyone knew that was coming. NVAX based
> > machines as such started shipping in 1991 as well.
> >
> >> I would have thought that it was intended to compete on:
> >> * max CPU in a single box
> >
> > NVAX does it better.
> >
> >> * max RAM in a single box
> >
> > NVAX does it better.
> >
> >> * max IO capacity in a single box
> >
> > In a single box, I think they come out even. The 9000 had massively more
> > I/O capacity, if you look at the full system. But that's a lot of boxes..
> >
> > And DEC was also pushing for clusters, and had been for quite a while,
> > where the capacity of a single machine wasn't the main point.
> >
> >> And just maybe its fast demise was also due to the fact that the
> >> mainframe market was moving to a single architecture (IBM mainframe
> >> with IBM, Amdahl and Hitachi as vendors).
> >
> > Possibly, but I wouldn't think so. There were plenty of DEC customers
> > looking for faster VAXen. And the VAX market was still strong at that time,
> > athough the Alpha was just about coming in as well.
> >
> > Johnny
> >
> If I'm not completaly wrong, the Swedish weather service (SMHI) had one
> VAX 9000 once. I seam to remember seeing it behind bars and fences at a
> visit to the SMHI HQ in Norrköping/Sweden. No one but DEC people was
> allowed to touch it. This was around the same time as a scandal with
> a 11/782 (I think) that was going to ship to Soviet.
>
> I also found this note from 1994:
> "1994-01-01, Over the last four years, CERN has progressively converted its
> central batch production facilities from classic mainframe platforms (Cray
> XMP, IBM, ESA, Vax 9000) to distributed RISC based facilities..."

I was on special assignment in Toronto in 1988 so recall a trade show where everyone was buzzing about the shift from CISC to RISC (so a lot of companies were jumping to Sun Microsystems). I recall being handed a document by a Sun rep comparing the price of memory between DEC and Sun, as well as overall bang vs buck. This might have been one-sided marketing but the DEC stuff was a lot more expensive in every category.

Alpha had been out for a while but we didn't buy any until 1996 (that was a pair of 4100s). One look at the industry standard motherboard slots was proof that DEC had finally got the message. But it was too little, and too late.

Neil Rieck
Waterloo, Ontario, Canada.
https://neilrieck.net
https://neilrieck.net/OpenVMS.html