[Beowulf] SGI and Sun: In Memoriam
Robert G. Brown
rgb at phy.duke.edu
Wed Apr 1 07:03:07 PDT 2009
On Wed, 1 Apr 2009, Kilian CAVALOTTI wrote:
> On Wednesday 01 April 2009 13:58:22 John Hearns wrote:
>> Not an April Fools.
> Whoa. That, and IBM eating Sun, the market is really shrinking...
Hijacking the thread not to argue per se but to write an elegy...
The market for high margin workstations and servers isn't just shrinking
-- it is long, slow collapsing, continuing a process that has been
inexorable and inevitable for just about exactly 20 years now.
SGI has been a zombie for a decade -- they were a company whose original
business plan evaporated out from under them as the cost of OTC graphics
hardware and associated software library support plummeted from the
"we're the only game in town" of the early 90's to the "My gaming PC
comes with state of the art graphics by default" of today. Their
secondary business plan -- supercomputing -- was destroyed by (gulp!)
us. Yep, it's our fault.
What are you going to buy, a $60,000 SGI 2x0 series refrigerator chassis
outfitted with $25,000 MIPS daughterboards to the tune of $160K or 16
Sparcstations that are twice as fast in aggregate and cover 16 desktops
(this would be an early 90's decision). Bear in mind that you can buy a
sparcstation a year for the software maintenance cost alone of the 2x0
series box, take your time...
The construction of the original PPro-based PC/Linux cluster a few years
later (1996) simply cut the cost of the standalone boxes in two (the P5
based systems just didn't have good enough floating point to properly
compete on peak even thought they did OK on aggregate) while SGI was
still engaged in high margin sales and dreaming of -- something -- was
then slow poison. They never managed to reinvent themselves. They
couldn't, as long as they persisted as a high-margin sales outfit in a
commodity sales world. If they'd gone cheap and embraced their doom,
they might have survived, but they continued to pursue an empty dream.
Sun is an even sadder tale. Sun had a brief, shining chance to own the
Universe, to become Microsoft (and crush both Microsoft and IBM), to
mostly crush the linux revolution before it got started (or at the very
least to alter its trajectory into something entirely different). In
1988 Sun released the Sun 386i -- I know, because I was one of the
original owners. This was way early in the lifetime of the PC. The 386
was the first Intel CPU that could operate in a flat memory model, and
with an 80387 coprocessor (later combined on a single chip in the 486
series and beyond) it had "respectable" numerical performance -- much
better than the Sun 3, not as good as SGI/MIPS or Sparc. The 20 MHz
386i Roadrunner was still way expensive -- mostly because of its 91 MB
hard drive, its enormous 4 MB or 8 MB of memory, its hi-resolution frame
buffer, its onboard Ethernet interface, and a custom motherboard with
proprietary slots for the memory and frame buffer -- but that's not the
The point is that Sun >>had a fully functional Intel Unix in 1988!<<
Alas, Sun persisted in the blind view that PCs were never going to
compete with Sparc, with MIPs, with DEC Ultrix boxes, with IBM AIX in
the workstation world. Even the Roadrunner was an afterthought,
carefully priced to not compete with Sparc and the project was abandoned
and killed in 1990 without actually releasing a 486i version.
Not content with owning one Intel Unix, Sun acquired a second one that
had failed to quite take off in the early 90's. They had not only all
of the technology and source for SunOS on Intel (including a lot of the
requisite drivers) they had a second set of sources for Intel Unix on
commodity systems with drivers for same. SunOS supported two kinds of
windowing (really three) -- Sunview, X11R3, and what was it called,
Openview or the like that was basically >>display postscript<<. Yep,
think scalable fonts onscreen, basically macintosh technology instead of
the #!@&^ unscalable pixel graphics in X that was then and continues to
be now an incredible, unbelievable, pain in the pitootie every single
bloody time screen resolution jumps a notch (he says typing on a
1920x1200 display wearing his reading glasses because the X fonts one
pixel wide are now so faint and tiny as to be all but unreadable).
Sun thus stood on the top of the mountain in 1990 -- the overwhelmingly
dominant workstation manufacturer, with the world's hands down best Unix
-- nobody who worked with a mix of the Unices then available would ever
argue with this, as AIX sucked, Ultrix sucked less but still sucked (and
the DEC workstations sucked), Irix sucked (although the SGI workstations
still didn't suck), and then where do you go? NeXT OS had potential but
it was blown by Jobs (sorry) when he had to go and make its back end all
proprietary and dysfunctional broken sort-of-Unix (die netinfo die).
Motorola Unix never really got off the ground. Apple's OS was a joke
and only available on tiny little "toy" boxes at enormous relative cost.
And then, off to the side, there was Microsoft and IBM.
At this point the writing was oh-so-clearly on the wall. The highest
end Intel OTC systems were maybe 1/3 of the price of the mainline Sun
Sparc boxes. The latter still had an edge in price performance --
seriously, right up to somewhere in the 1993-1995 range -- but for
>>desktop workstation usage<< as opposed to raw computation the gap was
>>closed<< by 1992 as the 486 really came of age and the inexorable
progress of Moore's Law took its differential toll, ever shrinking the
gap in raw performance. After all, for a desktop workstation one
doesn't, actually, need an enormously fast processor. 50 MHz is plenty
(that's 2.5x as fast as my already perfectly adequate 386i, more if one
accounts for faster memory and buses, more/cheaper memory, the emergence
of faster disks and scsi cards for the PC).
IBM and Microsoft were working on the infamous mix of Windows, NT, and
OS/2. Unknown to them all, a young student was working on his own
version of Unix for the Intel OTC architecture (although if they'd known
of it they would have scoffed, a silly thing to do given that >>all<< of
the main players in the PC software world or Unix workstation software
world got their start in metaphorical or literal garages).
At >>any time<< from 1990 through to 1995, Sun could have taken over the
world by simply releasing a fully supported version of SunOS for the x86
family for $50. That's all it would have taken. I mean, what are you
going to use? NT? OS/2? Windows 3.x? Or SunOS, especially if you
throw in a DOS loader so that users can run their old PC games and Lotus
spreadsheets until they buy the new SunOS versions of them.
At that time all the open source software in the Universe was built on
SunOS. Who thought that the community could actually support an actual
operating system and all of those device drivers? Truthfully, it
>>couldn't<< -- it took close to a decade before Linux was able to do it
and it is still way too spotty on bleeding edge new hardware or hardware
with a "proprietary" API and a linux-hostile manufacturer. I, and many,
many others, would have been >>thrilled<< to pay Sun $50, or even $100,
for real Unix on cheap OTC Intel boxes, and with Sun's enormous
marketing clout and an emerging Intel-based server market Microsoft,
Santa Cruz, Novell, all of the dumb and proprietary networks -- they
would have vanished like mist in the Sun.
Sure, in five or six years this would have killed Sun's Sparc business
(maybe -- it would have benefited too from the flood of software
companies writing PORTABLE code for SUNOS that would compile across the
two architectures) but >>nothing could save that<<. This is something
Sun never ever understood, right up into the 2000s. However, Sun as a
company would have survived virtually unchallenged and they might even
have kept Sparc alive LONGER, as there is an actual point to having a
high-end server architecture as long as there is software portability
and compatibility across the hardware boundary.
I remember telling Sun reps that this is what they should do in 1993, in
1995 (suggesting that they make then-new Slowaris of Evil die die die at
the same time), and in 2002 at a roadmap talk at Duke. By the latter,
it was clear that Sun's roadmap was a road to nowhere, but try telling
them that! All I heard from them was that they were sitting on a bank
of 6 or 7 billion dollars and had a plan; they were simply clueless as I
pointed out that the cost/benefit of linux based servers and linux based
commodity clusters and linux based workstations was an easy 3 to 1
greater than their persistent, damnable, high margin sales. Their reply
was only that they'd "give" us Solaris for Intels as WE were educational
users. Clueless... and an easy decade too late.
The big question is: Do the new owners "get it"? IBM gets it with
Linux, I think, and I myself will cheerfully buy IBM servers for places
that need tier 1 hardware that just doesn't break. IBM has a long,
smoldering war going on with Microsoft (their OS/2 memory is long, long
indeed). Apple has actually drunk half of the kool-ade they should have
drunk back in the early 90's (when they, too, could have taken over the
world in so very many ways). They're doing what Sun could have done in
the 90's -- supporting a semi-proprietary close to commodity hardware
platform by providing commercial support for the core OS only. But it
is a thin-margin Universe now as it has always been in the PC world, and
IBM and Apple and all the rest of the big boys never learned to live
lean on a consumer-driven diet.
So let us all bid adieu to two more of the ancient giants, killed by
their own lack of perception and market vision more than anything else,
killed by a refusal to see the world as it is and adapt to it unstead of
holding onto a broken dream of enormous profits not backed by the
delivery of anything like competitive or comparable value as the world
changed faster than they could.
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
Robert G. Brown http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567 Fax: 919-660-2525 email:rgb at phy.duke.edu
More information about the Beowulf