[Info-vax] Did Ken Olsen kill Alpha?

[glen herrmannsfeldt]

Neil Rieck <n.rieck at sympatico.ca> wrote:

(snip)
> I agree with almost everything you posted here except the VAX9000
> comment. In the book "DEC is dead, long live DEC" (which is a post
> mortem based upon employee interviews, there is a lot of input from
> Gordon Bell (he actually wrote all of Appendix "E") where he makes
> good arguments as to why computer architectures are only good for ten
> years or so, and that the time was right for DEC to move from 32-bit
> CISC (VAX) to 64-bit RISC (Alpha), just as DEC had done 10 years
> earlier in 1977 when they moved from 16-bit PDP (supporting physical
> memory) to 32-bit VAX (supporting virtual memory).

OK, for comparison look at what IBM has done, from S/360 to S/370,
to XA/370 to ESA/370 to ESA/390, and finally z/Architecture.
Those include going from 24 to 31 to 64 bit addressing, and
general registers from 32 to 64 bits.

So, six in almost 50 years, so about right for 10 years each, but
they maintain backward compatability at the user level all the
way through.  Programs compiled and linked on OS/360 over 40 years
ago will run without change on z/OS.  (I believe the PL/I compiler
has been tested, maybe other compilers, too.)

Now, part of VAX was compatibility mode, to run PDP-11 code, 
but that was dropped after not so long.  

Next read the "Foreward" in the "VAX Architecture Reference 
Manual" by Timothy E. Leonard.  Written in 1986, so about 10 years
after VAX started, he compares VAX to IBM S/360/370 and ends the
first paragraph with:

  "And though work on the first machines began in 1975, we expect
   VAX computers to remain the backbone of Digital's product 
   offerings for many years into the future."

Now, maybe IBM was lucky, in that S/360 was easier to extend
than VAX.  (Or maybe that was good planning from the beginning.)

Looking back, it is only very recently that virtual (task) 
address space really needed to be over 4GB, but VAX doesn't
really allow for that.  With the P0 and P1 distinction, one
is really stuck at 1GB or maybe 2GB.  It should have been
possible to fix this, though, for user programs.  Changes would
have been made to system space, and the OS, but could have
allowed for up to 2GB or 4GB user address space.

Did they try to go to 64 bits too early?  Before the demand,
and need, was there?  

> But after the death of Olsen's mentors, Olsen turned to his VPs for
> advice. Many of these VPs represented sales and claimed to have talked
> to customers demanding VAX 9000. But VAX 9000 flopped when it hit the
> market place. Did customers lie to DEC sales, or had DEC salespeople
> deluded themselves into believing there was customer demand? Not sure
> of the answer but DEC was never able to recover the development costs.

(snip)

> Speculation: IBM had huge financial problems between the late 1980s
> and early 1990s. I'll bet there were some people at DEC who believed
> that DEC would survive the 1990s while IBM might fail. These people
> could have made life better for both companies by having DEC
> manufacture their chip in an IBM fab, but this would be like throwing
> IBM a life line.

-- glen