[Info-vax] Out with Hurd, in with OpenVMS

[John Wallace]

On Aug 23, 1:24 am, Michael Kraemer <M.Krae... at gsi.de> wrote:
> Bill Gunshannon schrieb:
>
> > In article <i4qeso$ebq$0... at news.t-online.com>,
> >    Michael Kraemer <M.Krae... at gsi.de> writes:
>
> >>He/they tried at least three times to enter the high volume market,
> >>in the very beginning with the PCish alpha's,
> >>around 1995/96 with the Alpha PC (sp.?) and somewhere inbetween
> >>with he AXPvme stuff. None of that panned out.
>
> > And why do you think that was?  We had a Professor here who actually
> > bought one for a specific research project.  Imagine his surprise to
> > find out there was no software for it beyond the OS.  Well, there was
> > Linux, but not  much commercial to run on that.
>
> That's it, exactly. It's not the marketing,
> it's the product itself which hasn't enough
> incentives to justify the significantly higher price tag.
>
>
>
> >>But for a consistent strategy one *had* to reserve
> >>fab capacities. Just imagine the noise if it had turned
> >>out the other way: Alpha boxes in high demand but cannot
> >>be delivered due to CPUs in short supply.
>
> > It's all about management.  In a pinch you get the chips fabbed by someone
> > else until you can ramp up your capabilities.
>
> If it were that easy, why did DEC build a fab on their own at all?
> Would have been much less hassle and cheaper to let someone else do it,
> just like Sparc and Mips did.
>
>
>
> >>Mass market via VMS is just wishful thinking.
> >>I can't imagine typical PC customer falling in love
> >>with directories in square brackets.
>
> > Funny, that didn't slow down the rapid advance of Linux and BSD.
>
> Linux and BSD have "directories in square brackets"?
> What was the name of the marketing firm again
> who create all the ads for Linus Torvalds?
> (Must have been missing something in the past few years ...)
>
> > And,
> > remember, there was also Windows NT for the Alpha.
>
> That's what the hardware ran I was referring to.
> Except AXPvme, which ran VxWorks.
> A military project, btw, so you should know.
>
> > Not to mention that
> > by the time Alpha made the scene there already was a good GUI environment
> > for VMS.  The only things missing were software and marketing.
>
> Software? yes, that was missing, but marketing?
> As said, the Vobis Alpha PCs were exposed the same
> way as the other consumer PCs of that German
> retailer.

"Would have been much less hassle and cheaper to let someone else do
it,
just like Sparc and Mips did."

Which is why history records that Mitsubishi, Samsung, (and others?)
had Alpha technology licences (design and fab in Samsung's case).
Samsung even actually used their licence to build and sell things.

History also records that if you want to make the most out of a given
generation of fab, you have the fab's process people working as
closely as possible with the chip's design people, so they know just
what the fab can really do for the chip. A bit like Intel do. On the
other hand, AMD and ARM successfully do without fabs, and to a lesser
extent so did MIPS and SPARC.

Intel's other argument is that it's more profitable to build your own
fab, but the recent Dell/Intel court case at the US DoJ shows a
different possible reason for Intel/Dell profits. Go read about it.

Of course, as I mentioned earlier, there's more to system design than
the CPU itself. There's usually the bus interface chip(s), memory
controller chip(s), etc. And if it's a workstation there'll be posh
graphics of some kind. Someone needs to fab those too. Most of the
Alpha support chips of this nature were designed and fabbed in-house.
There was at least one well known exception, the Pyxis IO chip in the
PWS/Alpha family, which I believe was DEC design fabbed by a well
known name. Readers with working memories may remember this chip had a
problem, the workaround for which was (iirc) to disable DMA on the
IDE. Coincidence? Where cheap commercial alternatives were
appropriate, they were used - the usual SuperIO chip in a low end
alpha was a commodity chip, no point re-inventing that particular
wheel.

I don't know what the deal was with the DEC-designed (DEC-fabbed?)
Tulip network interface chip (21040) and its follow on but they show
up these days with various folks' logos on them, and have done so for
some considerable time.

"Except AXPvme, which ran VxWorks. A military project, btw, so you
should know. "

There was more to VME on Alpha than military although obviously
defence stuff was a large part of the VME market in general. CSS was a
part of DEC that got a lot of revenue from military stuff, though
whether it was always as much as the books appeared to show is
debatable. There's a book for someone to write about some of the
things that went on in CSS.

There was at least one German outfit (Atlas Elektronik GmbH?) that
designed and built their own VME/Alpha system(s) but I can't remember
the details or easily find a reference.

VMS was not the intended market for the VME kit, VxWorks and Tru64
were, although some customers in some markets would have loved to have
a VME-native Alpha card running VMS, rather than the hassle of
integrating a PCI-based (or, earlier, TURBOchannel-based) Alpha with a
VME adapter from DEC or elsewhere.

"the Vobis Alpha PCs were exposed the same way as the other consumer
PCs of that German retailer."

Maybe they were, but Vobis were largely a general PC builder and the
general PC market quickly found out that although MS had told Palmer
they were supporting non-x86 systems, MS didn't actually deliver what
they promised (who else remembers the Advanced RISC Computing
consortium? NT used to run on MIPS and PowerPC too).

In the NT market sectors where DEC focused on Alpha's strengths
(performance and price/performance) and where the wide availablity of
commodity software was irrelevant because there were a limited number
of crucial vendors whose support was all that mattered, Alpha was
actually doing OK (especially when you could get working kit in front
of prospective customers). The same story that worked in DTP and pre-
press for Apple; there are certain sectors where you don't need the
full Windows environment if you've got sufficient differentiation. HPC
would be another such sector. Marketing Alpha into the commodity
market was a nice idea especially as an awareness builder, but
expecting it to work with only half-baked support from MS was always
going to lead to trouble.

The other strength of Alpha was VMS but promoting that was anathema
inside most of DEC because "the market" had been rambling on about
"open systems" (but when it came to buying, they bought "cheap"
instead). Today, people buying VMS on IA64 are generally buying it
because they want VMS, they're buying it *despite* IA64 not because of
IA64 (exceptions currently apply at the really really high end, but a
Proliant-class box can do most of what most of the VMS market needs).

There's more to say but I need to be elsewhere.