[Beowulf] SGI and Sun: In Memoriam

Robert G. Brown rgb at phy.duke.edu
Thu Apr 2 09:33:47 EDT 2009


On Thu, 2 Apr 2009, Kilian CAVALOTTI wrote:

> On Wednesday 01 April 2009 17:48:56 you wrote:
>> A point I
>> didn't make in my blog post on this stuff is that the academic side,
>> while a nice market to play in, isn't terribly profitable.  You have to
>> give "killer" discounts to play in many cases.  Which in the case of a
>> low margin commodity business, is, in a real sense, a killer.
>
> But aren't those killer discounts paid back in image building and
> advertisement savings? It would make sense for companies to invest in some
> academic partnerships/projects, to make the buzz, and build up an image of
> performance, reliability and trust. Something along the lines of: "see, this
> prestigious university/institution is using that company's product line, and
> they must know what they're talking about, so I should do it too."

Not really, not "in general".  For certain fairly specific products,
sure.  The question is, who is going to be influenced by their "free"
college experience to make an "expensive" long term purchase decision?

For Microsoft-anything, there is an enormous post-college market for
Microsoft-everything.  So Microsoft usually gives Universities and
schools a big break.  Raise up little Windows-heads, and they'll
presumably be loyal when they reach corporate america or just go to Best
Buy to buy systems of their own.

This is impossible for Sun-anything.  There IS no such thing as
"consumer Solaris" or over the counter machines with Solaris on them for
use at home or work.  This is an important counterpoint to Linux -- even
though for most of its life Linux has been "expert friendly" on the
desktop compared to preinstalled Windows on a box from whoever, it has
ALWAYS been a desktop system first and everything else second.  Linus
has rejected many, many alterations to the kernel over many, many years
now on the basis that they'd negatively impact desktop performance
(although that's not much of an issue anymore on modern hardware and I'm
guessing that policy is de facto more relaxed at this point).

To take Duke as an example, we have a campus mirror for Centos and
Fedora and have a (currently indifferently) supported "Duke
distribution" of both that is basically a few add on yum repos with site
license or Duke restricted stuff, need a Duke IP number to access.
Many, many users -- students, faculty, departments -- here and elsewhere
use this site to simply pop linux onto their systems, be those systems
desktops, servers, laptops, whatever.  We even have a vmware site
license at this point, which is making Win under Lin (using our Windows
site license) a knee-jerk sensible choice for many, although dual boot
is still favored by gamers.

In the process, users lose their fear of Linux (and learn of its utility
for providing access to e.g. the best, free, programming tools, to
firefox and mail, to open office, to most of what one needs on a desktop
system and usually have to pay for on a Windows box.

SURE we can could put Solaris on systems for free as well.  But how?
And why would we want to do such a thing?

Does Solaris support dual boot?  I don't know, but I sort of doubt it.
Only if I install grub from linux or use MS's multiboot manager -- maybe
then.

Does Solaris support a free network install mode?  To install linux on
campus one at MOST downloads a boot image and boots it (from a USB key
or CD) although in many (tecnical/science) departments booting on a wire
gives one a set of PXE boot images to choose from for an install and in
some (like physics) one PXE boot into a system-specific custom generated
kickstart.  Most students can manage this with at most a bit of help,
and of course it is obviously free and fairly well publicized.

Nobody outside of the computer science department and a few techno-heads
like myself even knows that Solaris is available for free.  There is no
easy path facilitating its installation anywhere.  No server.  No PXE.
No support.  I'm an ex-Sun head (very ex at this point, but ex) and I
have no idea how to go about getting Solaris to even try to install
under vmware just to play with, and have no idea why I'd ever want to do
such a thing.  I'm guessing that I could visit Sun's website and fill in
a form swearing I'm University affiliated, or maybe get it through OIT
or the CPS department, but just figuring out where it is and how to get
it would take longer than actually installing Linux the hard way, let
alone via kickstart.

So IF Sun's plan was to use free distribution as a loss-leader, it was
hopelessly executed.  But let's move on, back to the question of why
would I WANT to install Solaris?  Is there one, single thing that I can
expect to do on Solaris that I can't do better -- far better -- in
Linux?

Let's see.  Solaris is Unix.  Linux is Unix.  Solaris, with some work,
comes with some set of packages that hopefully includes a compiler,
although as already noted at one time they unbundled the compiler to
their everlasting dismay.  Unix without cc is like a day without
sunshine, just plain stupid.  REALLY stupid NOW, with infinite room for
the OS image compared to nearly all possible boot/execution media.  I'm
guessing that it comes with X, some sort of window manager, and a
handful of applications, at least I hope it comes with some
applications.

Linux -- well let's see.  15,000 packages in Fedora 10, who knows how
many in current Debian (25,000?  30,000?).  You want to be able to take
a PDF and extract page three and convert it into a PNM image?  As they
say in Mac-land (where it is, of course, untrue) "there's an app for
that" (taking the liberty of calling a short pipe of commands an app)
and app is a yum install away if it isn't already installed by default.
Want to remove red-eye from photographs and print them?  There's an app
for that.  Want to monitor a cluster of linux systems?  There's an app
for that.  Want to test random number generators with a tool integrated
with R? There's an app for that -- apt-get and it can be on your system,
for free, in a few seconds.  Security concerns?  No problem -- linux
updates these days all day long.  The window for a successful exploit
from the time of discovery has shrunk from 24-36 hours three or four
hours ago to as little as 3-4 hours today.  Conficker that.

APPLICATIONS drive operating systems.  This has been true since
operating systems were invented.  Nobody boots a system to stare at a
command prompt.  So what are the applications available for Solaris and
how do they help differentiate it?  How easy/hard are they to obtain and
install?  Who supports them?

The answer is: the same list of GPL-alike apps that are available to all
of the OSOS's, only way, way shorter, plus the usual mix of commercial
apps that are of interest only to a very limited community of users.  We
know who supports the open source apps and while Sun does contribute
(open office, SGE) it is still not, really Sun.

Here's the kicker.  Sun (or by proxy now, IBM) could even now, today,
become a "player".  How?  Dump Solaris.  Dump Sparc (please!)

Embrace the penguin.  And if you must compete on the CPU front do it
with an Intel compatible general purpose CPU, not a proprietary, single
source, non-commodity (and other flaws not worth mentioning) failed
architecture that everybody but Sun has known was toast for over a
decade now.

Think about it.  "Open Solaris" or not, Solaris is pointless.  How many
Unixoid environments with irrelevant differences do we need?  We already
have too many.  A few have relevant differences (and I'm the last person
to argue that there isn't advantage in innovation and competition) but
put those innovations where they belong in an open source world, not
into an operating system that -- "open" or not -- is PROPRIETARY.

Now imagine Sun Linux.  Oooh, has a ring to it, doesn't it.  Sun is one
of two or three companies that could credibly compete in the Linux
marketplace in this way from day one.  The only others I can think of
offhand are Apple, IBM, and Microsoft (although hell will have to
descend to liquid nitrogen temperatures or below for the latter to be
credibly likely) and of these, Sun actually has the best cachet and
could conceivably pull off actual market dominance.

Imagine Sun contributing to the linux kernel.  I have the greatest
respect for Sun's coders -- some of the best in the world there (and I'm
sure a lot of them contribute to linux already, officially or
unofficially).  Imagine Sun cutting deals with hardware and software
companies to open up sources for hardware drivers and to guarantee
support for software ports.  (Imagine IBM pitching in here as well,
especially on the server marketplace where IBM/Intel hardware is a
dominant force).  They could give Red Hat a run for its money from day
one.

IMO, the "RISC vs CISC" wars of the 90's are long since over and CISC
(usually in the form of a CISC shell over a RISC-like engine or other
hybrid approaches) has one.  It one not because of the theoretical
arguments that dominated the 90's but because silicon itself "voted" --
it turned out to be necessary to go multicore instead of continue to try
to push clock; single CPU systems suddenly became "clusters" (or at
least de facto SMP systems).  If there's room to put four entire cores
on a piece of silicon with an enormous L2 cache, and one is clock
limited on that system for other reasons (like heat dissipation
capabilities) the CISC vs RISC argument becomes irrelevant to
performance because real estate needed for a translator from CISC to
whatever the CPU "really" uses at the ALU level stops being all that
critical a resource.

The original argument (for those who came in late, since articles on
this subject all but dried up by the early 2000's because nobody cares
any more) was that with RISC you can put more, simpler, ALUs on a chip
by not using real estate for complex instruction translation, and make
better use of those ALUs with smart compilers, which of course produced
larger code using fewer, more elementary commands, that in aggregate
make a single instruction on a RISC system.  This in turn requires much
better than average compilers and libraries and uses up more RAM in
operation but is supposed to give you a faster CPU, all things being
equal.

The problem is that they have never been equal and will never be equal.
All that matters to the consumer is price/performance, and for better or
worse, the RISC-iest processors have never been beloved by the
marketplace because a) their price/performance has sucked at the
hardware level; and b) nobody has written lots of software for them at
the software level EXCEPT the one company that owned or was the
major/only consumer of the CPUs.  CISC code is cheap to build and sort
of optimize, CISC CPUs are mostly CISC translators on top of RISC-y back
ends that are pre-optimized for the most likely execution pathways
anyway, so at most one gets a rare boost from a RISC for some special
application.  This isn't to disrespect RISC -- I was very fond of SPARC
for many years, but honestly I didn't really care about SPARC per se,
what I cared about was "make myapplication" doing a perfect compile and
producing a result as cheaply and quickly as possible, and the ability
to find applications I could just drop onto the system and run that were
built by Other People, for free (saving me a ton of money and time that
vastly, vastly outweighed the CB of the argument for nearly all possible
uses of the system).

That's why nobody cares.  Power PC a good CPU?  Sure, why not?  But
unless it is two times better than Intel AT CONSTANT OVERALL COST and
unless there is just as much prebuilt software and a hundred vendors
using it as the basis of commodity platforms that are totally vanilla
and interoperable, it won't touch Intel/AMD.  To touch them it has to
become them -- be implemented by hundreds of vendors (not just Apple
and/or IBM), provide a vast wealth of prebuilt software, be cheap AND
STILL be faster than either one.

So here's an idea for Sun -- nobody loves your SPARC, no matter how
technically good it is.  Nobody will ever love SPARC, ever again.  It is
done.  Finished.  Forget it.  Gone.  This Parrot is Expired!

So take the damn parrot apart.  There's some great technology in there.
Put an Intel/CISC translator onboard, and feed your SPARC-y back end
utterly compatible code.  Sell it in direct in your face competition
with Intel -- don't even think of keeping it proprietary and for
in-house use only.  Stop thinking of yourself as a hardware company
selling computers.  That was Apple's problem -- they were a software
company that thought it was a computer hardware company.  Once they
(finally) realized that the hardware was commodity and the software was
key, once they focused on OTHER hardware like ipods and iphones and made
an intel-based Mac, they actually started to make money and survived!
You can do the same thing, differently.  You've GOT the silicon foundry,
just use it to make something that you can actually sell to somebody
other than a direct market of rather stupid end-users who really should
have bought Intel or AMD servers and required all of your wiles to
decide otherwise.

That really is a scary thought.  Think of all the companies that would
absolutely hate it if Sun pursued this strategy.  Intel would hate it;
they tolerate loss of business to AMD, but they're still in the driver's
seat in the world CPU marketplace.  Sun AND AMD, both making competing
chips, quite possibly both making competing chips that are faster AND
cheaper than Intel's -- Houston, that would be a problem.  AMD would
hate it too -- it would be enough to kill them off, although it could
also weaken Intel enough to actually improve their position and create
an actual free market where they are a rational choice.  Red Hat would
REALLY hate it.  Right now they are the venerable old man of the
commercial Linux world, but Sun is positively antique and has enormous
cachet in the computing professional industry in spite of their
silliness and blindness of the past fifteen post-linux/intel years.

I manage an Intel system running Solaris in a small medical practice.
It is running at this very moment, on a cheap, over the counter Intel
box with 4 GB disks.  Yes, it is that old.  It hasn't been updated in a
decade or more.  I'm sure that it could be cracked in an instant, if you
could find a cracker that actually remembered the exploits of that long
ago.  This system has run, continuously, for twelve years.  It never
crashes (which is a damn good thing, because it is broken enough at this
point that I have to bring up its network by hand from the console).
Its commercial software package (sole reason for its existence) still
works, still serves the data it contains.  THAT'S the kind of reputation
Sun can bring to the Linux party -- Linux can do this sort of thing too,
but who can beat Sun in terms of overall Unix-expertness?  Not a whole
lot of companies can even come close.

Microsoft would hate it, because it provides yet another set of
experienced salespeople pushing the Penguin, an Othello-like flip of all
those Solaris boxes into Linux instead of Windows 200x server.  Sun has
the leverage to fix the eternal hardware driver issue.  Sun would bring
along at least some commercial applications and increase this space.
Sun COULD even join with IBM and introduce consumer linux big time,
complete with TV advertisements.  IBM has never forgotten who made
Microsoft into Microsoft and who got screwed by Microsoft a mere decade
later.  With Open Office onboard for free, with margins that Microsoft
can never compete with no matter what they do (short of embracing the
Penguin themselves and it isn't that cold outside yet), with businesses
looking for ways to save all that money they are currently spending on
Microsoft's enormously expensive software suite, with conficker and many
other perpetual embarrassments and increasing pressure from Europe,
China, Russia (all sick of paying monopoly prices for a broken,
insecure, mess) all it would take is the NAMES IBM and Sun to create
instant long-term confidence.

Ah, dreams.  Fantasies, perhaps, but interesting ones.  Heck, let's
bring SGI back to life in similar fashion -- they too have some damn
good coders, they've got the name and cachet, they just need the large
market that would appear if the world ever turned the corner on
Microsoft.

> That seems to me like a short-term profitability loss, but a long-term gain
> since it could drive more sales.

Sales to who?  The only users of Solaris systems in Universities are
computer science types (at least around here) and there aren't a lot of
even them.  The big markets that justify the loss-leader approach are
consumer, corporate, gaming, even large scale research.  Only the latter
gets any boost at all from such a plan, as the corporate share has been
shrinking (lost to lin and win) in SPITE of it.  And loss-leading cannot
compete with advantageous CBA.  The bottom line is the bottom line --
it has to be at least comparable to justify a free choice.  It hasn't
been comparable for years.

    rgb

>
> Cheers,
> -- 
> Kilian
> _______________________________________________
> 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


_______________________________________________
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf

-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.



More information about the Beowulf mailing list