[Beowulf] Re: vectors vs. loops
Robert G. Brown
rgb at phy.duke.edu
Wed May 4 11:49:39 PDT 2005
On Wed, 4 May 2005, Philippe Blaise wrote:
> >
> >Just as any market place it is ruled by price/performance. One could argue
> >that during the "vector machine" epoch there was an artificial market that
> >was due to the cold war. Performance at any price was an important
> >strategic issue.
> >
> >
> >
> You really think, that even today, the market place is ruled by price /
> performance ratio ?
> So why don't we use millions of slow and cheap cpus ?
> With this kind of argument we must stop the research programs, and try
> to transform all the
> scientific codes for embarrisingly parallel computing !
Um, we DO use millions of slow and cheap cpus. And most of the
important scientific codes have been transformed for embarrassingly
parallel computing -- it is what "Grids" are all about. The ones that
cannot be done EP up front have in many cases been parallelized the hard
way to run on network-based cluster architectures.
This is the beowulf list, remember? That's what it's all about!
And the reason the list STILL exists, some ten years after its
inception, and is as active today as it ever has been (which is so
active that it is a serious drain on the participants time) is because
the marketplace is ruled by price/performance, and cluster-based HPC is
a price/performance winner compared to nearly any alternative by a
factor of anywhere between 3 and 100 for "most" (that danged word
again:-) HPC applications. That is, nothing else comes close, sometimes
by such an embarrassingly large factor that even conservative heads of
corporate data centers have long since been convinced that this is the
right way to go and indeed very nearly the only game in town. Nowadays
the question is much more what KIND of cluster one gets and whether or
not it is COTS or big iron. Without wanting to get into any kind of war
over whether a current-gen big-iron SMP "supercomputers" (e.g. an Altix)
is or isn't really a "cluster" if they use e.g. cc-NUMA shared memory
rather than network-based memory passing per se as the IPC paradigm.
Not all "networks" are external, and not all clusters are COTS, which
fuzzes things at some point.
> More seriously, the market place seems mostly governed by big US
> companies and
> financial directors.
> Moreover, remember that some years ago NEC and Fujitsu supercomputers
> have been
> taxed by the US government. (and that was not during the cold war !).
> See for example
> http://www.freetrade.org/pubs/articles/cd-supercomputer.html
>
> Today everyone should be happy to see that some companies like Cray
> are trying to do better than HP or IBM in the linux cluster area, event
> it if it's not
> the Cray of yesterday as you said.
> I think that this is partially due to the "NEC earth simulator effect",
> and that linux clusters
> performances and functionalities are not fantastic for parallel jobs
> compare to a good old T3E.
I think that the cold-war/government support effect is far from
finished, myself. Supercomputers are a major component in weather
prediction, decryption and intelligence, weapons systems design
(including nuclear weapons), space research. "Super" computers are
increasingly integrated into weapons (systems) themselves or carried
aloft in rocket payloads (as the efforts of Jim Lux clearly
demonstrate).
Further, the fabled military-industrial complex is tightly coupled, with
all sorts of backscratching and sweetheart deals, some of which are
simply designed to keep a strategically important company afloat in
peacetime so they'll be there if a war should start. I'm sure we're all
aware of examples of this in the aerospace industry, for example.
I think this is one explanation for the persistence of the big-iron
supercomputer and even some "competition" in the marketplace for the
limited number of really big systems that are produced and sold.
> The price/perf ratio is too simplistic.
> The marketplace at a given instant means nothing.
Granted that things are not always about price/performance (which is why
linux has not completed its takeover of the world, for example).
Monopolies are all about taking over a marketplace by offering
short-term P/P wins to wipe out your competitors followed by cranking
the price to the limit the market will tolerate once that annoying
"competition" thing is gone. P/P is still important, though, and in the
LONG run tends to win as a strong P/P imbalance provides a competitive
advantage to those that take advantage of it. Linux has an "intrinsic"
P/P edge over (say) Windows that makes it more or less impossible to
wipe out by simply dropping margins for a while, so as time passes the
market share of Linux tends to gradually increase and barring a
nonlinear constraint (a patented or copyrighted technology that becomes
the cornerstone that perpetuates monopoly) will eventually take over the
world, as promised.
Clusters, on the other hand, have proven SO successful that they HAVE
taken over the world except for small, protected, costly but still
powerful island protectorates. Clusters are everyman's supercomputer --
you can have one at home (I do:-). They are "democratic" where big-iron
is still a monarchy and requires a sort of vendor-allegiance and ability
to pay high prices.
As for the marketplace at any given instant meaning nothing: Wow man,
that's kind of Zen don't you think? I mean, what DOES anything mean,
after all, especially in a snapshot taken at any given instant?
Whatever it MEANS, the marketplace is sure damn IMPORTANT to a lot of
people -- economists, corporate boards, consumers of all sorts and
levels seeking to fullfill their needs and desires at a bargain price.
The marketplace is the ultimate place where the dreamers and the doers
and the buyers all meet to do their business, each to their own
advantage.
I think Doug said it very well -- it costs billions of dollars to set up
a fabrication of a new modern CPU design, and only sales in the millions
in the mass market or in the thousands in a restricted and much more
expensive market can break even for the company involved. Most
companies, even cash rich and heavily capitalized companies like DEC and
IBM and Sun and Apple can afford about one mistake, sometimes two,
before the "marketplace" wipes them out, eaten alive by more successful
or wiser competitors. Where is DEC today? Apple only survives because
Microsoft in particular couldn't afford for them to go under while they
were in the middle of a major antitrust suit. Sun is working hard to
reinvent themselves while spending down a big pile of cash; SGI is still
recovering from THEIR forcable reinvention as a supercomputer company
when high end graphics adapters suddenly became cheap and plentiful. Of
the major players out there today, only IBM and Intel (in my opinion)
can afford to take a major risk and LOSE -- for everybody else the next
time they come up seriously short on next-gen technology they probably
go under.
Maybe the marketplace means life and death, both metaphorically and
literally, for whole companies, thousands of employees, millions of
stockholders. The money and lives that are risked are real, as are the
rewards, and the marketplace is as pitiless as any lion pulling down a
gazelle, as any gazelle that turns around and punches out the eye of the
lion with a sharp horn, as any buzzard that feasts on the remains of the
loser while the winner fattens beneath the summer sun. None of us can
escape from the marketplace -- we sell our labor and buy our daily bread
and cope as best we can with our finite means to realize our boundless
dreams.
That seems pretty Zen too...;-)
rgb
>
>
> Phil.
>
--
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