[ocaml-biz] The Hype Cycle

[Brian Hurt]

Long, mostly off topic, but with a brilliant save at the end to bring it 
back on topic.

On Tue, 31 Aug 2004, Brandon J. Van Every wrote:

> 
> Man do I personally hate Merced.  I put my early career into mastering
> the DEC Alpha, the best CPU that had shipped anywhere.  DEC sued Intel,
> DEC got money.  DEC became cash attractive and Compaq gobbled it up.
> DECcies lost faith in the Alpha, internally within the company.  We'd
> get lied to in employee meetings about how Compaq was going to "do
> something" with the Alpha license, not just kill it.  I left DEC before
> the Compaq buyout was completed.  I saw the writing on the wall.

I note with humor that it's only been in the last few *months* that the 
Alpha has (finally!) been knocked off it's throne as king of SpecFP:
http://www.aceshardware.com/SPECmine/index.jsp?b=2&s=0&v=1&if=0&ncf=1&nct=256&cpcf=1&cpct=2&mf=250&mt=3600&o=0&o=1
It's still a competitor.

Not bad for a dead chip.  I wonder what it would be doing if it was still 
being developed actively.

> 
> All these years later, those Intel assholes still don't have anything
> better than the x86 to offer.  But I have to give them credit where
> credit is due.  Intel had DEC whipped on fab, that's how they put the
> nails in our coffin.  And DEC couldn't market its way out of the paper
> bag.  Many sad lessons of disillusionment for a young techie.  I wanted
> to believe in something.  I found that corporations cannot, under any
> circumstances, be believed in.

The problem is, I think, that the IBM monopoly is alive and well, and 
currently called the Wintel Duopoly.  If you think the Alpha kicked ass, 
take a long, hard look at the CDC-6600.  Among other things, that machine 
made Seymore Cray.

The advantage to the Wintel Duopoly over the IBM monopoly is that the 
Duopoly has no experience in the long-term maintainance of a monopoly.  
Remember, IBM was founded in 1896- 50 years before the advent of 
computers- explicitly to be the punch card monopoly.  Yes, Virginia, both 
punch cards and IBM predate computers by quite a bit.  A variant of the 
Jaquard loom allowed basic sorting, tabulation, etc. operations to be 
performed on punch cards long before computers came onto the scene.  IBM 
only got into computers to protect it's monopoly on punch card machines 
and punch cards.

In almost 100 years of maintaining a monopoly, IBM only screwed up once.  
Unfortunately for IBM, once was enough.  It made a computer that it sold 
more than dozen of in which is did not make the CPU and write the OS- the 
IBM PC.  Opps.

Intel and Microsoft, on the other hand, are screwing up right and left.  
Microsoft is trying to act like a small growth company with $16 billion
dollars a year in revenue.  The stock options, the lure of becoming a 
Microsoft Millionaire, the 20%/yr growth curves- all these things are the 
hallmarks of a small company.  Unfortunately, a 20% growth in sales means 
they need to find $3.2 billion in *new* sales every years.  Next year, 
it'll be almost $4 billion in new sales.  You can't do it.  You can't 
invent, and completely dominate, a new $3 billion/year market every year 
like clockwork- there aren't that many markets.  So what does Microsoft 
do?  They raise prices.  About 20% a year.

Sooner or later, the customers *will* revolt.  And Microsoft will suddenly
discover that it has no friends.  If Windows still only cost $10/CPU for
the OEMs, and OTC list price was $70 but shop around, you can find it for
$30, Linux wouldn't be a threat (well, as much of a threat).

Intel fell into a different trap.  They've kept the lid on prices, but 
forgot to keep the competitors out.  IBM segmented the market into three 
basic categories- entry level, mid range, and high end.  And it had a 
policy of never letting a competitor gain a signifigant toe hold in any 
segment of the market.  It would sell computers at a serious loss, at 40% 
under what it cost IBM to make them, to keep competitors out.  Because 
losses in one segment could be recouped in other segments that weren't 
facing competition.  It was entry-level that they most often had to lose 
money on.

Intel has kept IBM's original three segments- consider Celeron, Pentium, 
and Xeon brands.  What are these, except entry level, mid range, and high 
end?  But between the 486SX and the Celeron, Intel abandoned the low end, 
because it wasn't profitable enough.  This was the Ur-screw-up they have 
yet to recover from.

AMD's break through was the K-6 chip, targeted primarily at the low end.  
It was nothing to write home about, but it was an OK performer, and it was 
real inexpensive.  And Intel's initial, paniced rush out of the cacheless 
Celeron didn't help things.

The K-6 gave AMD the money to develop the Athlon.  When AMD lapped the 
Athlon and beat Intel to 1GHz, Intel again had to panic, and screwed up (I 
beleive) even worse.  They pulled engineers off of Merced and threw them 
at the P4.  In addition to kneecapping Merced (Merced would have been a 
lot more impressive four-five years ago) it didn't help the P4.  I have no 
direct prove, but I have circumstantial evidence that Intel tried 
dramatically increasing the number of pipeline stages in the P4 late in 
the game, and ended up with unbalanced stages.  Exhibit A: the double 
pumped adder unit (making lemonaide out of lemons).  Exhibit B: the 
scaling problems they've had.  Exhibit C: the fact that they added another 
10 stages in the most recent revision.  But the point is: now Intel had 
competition in two of the three segements- both the Celeron and the 
Pentium were facing AMD competition.

But the Athlon set AMD up for the Opteron.  Which competes against the 
Xeon.  Now Intel is facing something that Intel never did- a single 
competitor (not a dozen competitors, a single competitor) with a 20%+ 
market share and the ability to compete with Intel across the board.  And 
maybe even go into places Intel never could (be real- how many 
32-processor x86 machines are really sold each year?  Compared to 
32-processor Power, Sparc, or PA-RISC machines?).  

And Intel is panicing yet again with the whole ia64 thing.  I wonder when 
people will wake up and realize AMD is no longer Intel compatible- Intel 
is AMD compatible.  Hint to Intel: you have the tech writers on staff.  
When stealing some one else's architecture, rewrite the manuals from 
scratch:
http://www.theregister.co.uk/2004/04/07/intel_64bit/

Even worse, take a look at the benchmarks.  Ignore the first person 
shooter games and the media encoding, and look at the webserver and 
especially the database performance numbers.  AMD is absolutely spanking 
Intel in performance in things people are willing to pay real money for 
performance in.

If I were AMD, I'd quite happily cede the gamer and media encoding markets 
to Intel, in exchange for the data base, web server, and business logic 
markets.

> 
> AMD Opteron's success will be sweet revenge though.  'Round the time I
> left, AMD and DEC were in an alliance / partnership regarding floating
> point technology.

Worse than that.  AMD cherry picked the Alpha hardware team to make the 
Athlon.  It has also cherry picked the MIPS hardware team.

But the point I think I'm making, in an incredibly long winded fasion, is 
that I don't think marketing can create opportunities.  Opportunities 
exist, or they don't, irregardless.  I don't think the Alpha died and AMD 
thrived because of marketing or the lack there of.  I think AMD was just 
lucky (finally), and got an opportunity.  DEC was lucky to, in the past- 
the PDP-1, PDP-8, PDP-11, and VAX all exploited opportunites.  They were 
the computers DEC was built on.  It's just was someone else's turn to be 
lucky.

Likewise, Intel and Microsoft didn't get to where they were solely by 
their own efforts.  IBM fumbled the ball, and created the opportunity for 
them to take over the monopoly.  Intel and Microsoft's fumbling are 
providing opportunities for AMD and Linux, respectively.

Java fumbled the ball as well- and there's an opportunity laying there for 
someone else to pick up- maybe Ocaml, maybe Python, maybe someone else.  
As a business logic language, Java was a roaring successs- and did what no 
other language has managed to do in 30 years, and finally actually kill 
Cobol.  As an applications level language, it's been an abysmal failure, 
especially as an open source applications language.  C++ is also entering 
it's trough of disillusionment (hopefully never to return), opening up an 
opportunity for some other language to move in.

-- 
"Usenet is like a herd of performing elephants with diarrhea -- massive,
difficult to redirect, awe-inspiring, entertaining, and a source of
mind-boggling amounts of excrement when you least expect it."
                                - Gene Spafford 
Brian