[Beowulf] SGI and Sun: In Memoriam

Robert G. Brown rgb at phy.duke.edu
Wed Apr 1 08:39:22 PDT 2009

On Wed, 1 Apr 2009, Joe Landman wrote:

> Robert G. Brown wrote:
>> 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,
> s/persisted/persists/
> This is a current religious mantra within Sun, that Solaris/Sparc will 
> (eventually) kill Linux and x86/x64.  Um.  No.  That game has been over for a 
> while now.

Precisely.  I don't know if it is a corporate culture thing, or if
they're taken into a room and conditioned with electroshock.  Otherwise
they are often intelligent, well spoken people who live in nice houses
and think that one day the world will come to their senses and resume
paying them 100% to 200% margins so they can stay there.  Maybe the
scientologists have gotten a hold of them...

> [...]
>> 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
> Hmm... I take issue with "Irix sucked" ... then again I am biased as I 
> learned Irix right after Unicos and Sunos.  Compared to either of those, Irix 
> (from a user perspective) definitely did not suck.

Um, as I said "SGI workstations didn't suck".  The workstation GUI was
just lovely, the programming interface was a bit dicey but useable, and
for graphics it was the only game in town.  My perspectives of the day
came from being a systems/network admin for a highly heterogeneous
network of Suns, SGIs, NeXTs, Apples, PCs, with a few groups stubbornly
stuck on Vaxen and one or two unixoid mavericks mixed in, plus
experiences gleaned on AIX, Ultrix, and several others.

SunOS may have been BSD based, but it was really, really easy to manage
and as I said, all the open source software in the world of the day
would just build on a Sun, because that's where it was developed.  It
required dedication, patience, and a fair bit of teeth-gritting to build
our full suite of "standard" software for the Irix, even more for NeXTs
etc (infinity for AIX and I just didn't try, less so for Ultrix which
was moderately SunOS-like in places although filled with proprietary
stuff and restrictions like their counting rsh users against their
login/license counters IIRC, where Sun didn't care if 100 people logged
into one of their boxes remotely -- they just didn't count.).
Administratively, the world looked like NIS and NFS, period, and NIS
worked perfectly on Suns (I started using it there back when it was
still Yellow Pages and yeah, they did invent it:-).  NFS ditto.  In
fact, they invented or were strongly involved in the invention of most
of the modern networking stack, and on Suns it all worked.  On Irices it
mostly worked, with some effort and hacks.  On NeXTs, grrrrrrrrr kill
kill die netinfo die and what was with the installed man pages referring
to nonexistent Unix commands...)

Perhaps "sucked" is too strong a word, but they were a much bigger
headache than the Suns.  Everything was a bigger headache than the suns.
Hell, 2/3 of our Suns were diskless -- Sun more or less invented the
diskless workstation and all the protocols that supported them IIRC.
Way cheap, just as fast on numerical code that would fit.  Irises made
good clients, but nobody I know would have founded a network on an Irix

>> 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
> Hmmm.... I was running MD on a desktop 486 machine that outpaced the Sun unit 
> (supposedly a faster box) in 93 or so.  Sun has rarely if ever had a fast 
> chip/box.

Sparcs outperformed Intel (usually by a factor of two or more) right up
to the Ultra.  The Pentium Pro was really the first chip that could hold
its own and even win against regular Sparc, but even then the Ultrasparc
was faster (and I did TONS of benchmarking then as I shopped and ran on
both and know this to be true).  The problem was that Ultrasparc was
only a BIT faster and cost roughly 3x as much, 4x if you got bleeding
edge.  As far as price/performance in aggregate flops, I'd have to let
Don address that one.  The original Pentium beowulf I think did end up
cheaper than one could ever have done with a stack of sun's most CBA
advantageous boxes, but I think it also took 2-4 times as many boxes to
get there, so for some classes of parallel code it was probably a loss
-- except Don invented channel bonded ethernet and bumped the Amdahlian
limit and maybe won there as well.

But with the PPro there was never any question.  200 MHz, dual CPU,
supported large memory and big disks -- a real workstation and it
benchmarked out beating current gen Sparc and was hot on the heels of
Ultra, I think reaching it in 1997 or 1998, don't remember exactly when
as by then I'd stopped even looking at Suns.  They were STILL 3x as
expensive and they ran Solaris, which really, really (I do mean really)
sucked right through when I truly stopped caring.

> Spot on.  A toast to two fallen giants.  I raise my cup of coffee to them 
> both.

Here, let me add a little spot of brandy to that cup.  They deserve at
least that... for it was upon their shoulders and out of their egg that
Linux was borne, as all the SunOS and Irix users in the Universe,
frustrated with high prices and incompatibility and stupid corporate
changes, leapt onto the Linux wagon in droves when the CBA started to
strongly favor it.


>>    rgb
> -- 
> Joseph Landman, Ph.D
> Founder and CEO
> Scalable Informatics LLC,
> email: landman at scalableinformatics.com
> web  : http://www.scalableinformatics.com
> phone: +1 734 786 8423
> fax  : +1 734 786 8452
> cell : +1 734 612 4615

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 mailing list