Readers offer their opinions on the recent debate raging in HPCwire over the future of high-performance computing in the United States. See last week's letters [385631] for more. letters from readers.
This week's reactions are below.
Tim,
Mr. Larson writes “HEC argues national security needs much better supercomputing; industry won't support supercomputing R&D; therefore, government should support what industry will not. Market failure is a fundamental premise in HEC's argument for government sponsored revitalization.”
I think the argument that industry won't support supercomputing R&D is fundamentally wrong. Industry can't afford to fund supercomputing R&D unless it leads to a viable product. Viable products must be self supporting. The failure is on the part of the supercomputing buyer, who expects to pay a commodity price for a custom product. To be fair, the cost of a custom product is well north of considerable.
The HECRTF focused on technical issues. I think we need a similar meeting to discuss the business aspects of supercomputing. Federal agencies, including OMB, need to have a clear understanding of manufacturing issues in order to address how best to fund the industry for endeavors that must span multiple years. Industry needs to understand how agencies respond to the President's Management Agenda and the requirement for measurable results.
Gary Wohl
NOAA/National Weather Service
—–
Dear Tim,
To me the arguments brought up both by the High End Crusader and by the High End Agnostic, as well as by Brian Larson, all sound hollow, because the current state of the HPC industry *is* the result of government meddling and not of free markets to begin with – if only by the virtue of the government being the major customer of the HPC industry, followed closely by companies that make things for the government (e.g., the defense and the aero-space industry), and institutions that are government sponsored or funded (e.g., national supercomputer centers and national laboratories).
In the days of the old government, their suppliers and their grantees used to buy Crays, Convexes, MasPars, Connection Machines, Paragons and similar exotic systems. This was a blatant act of meddling with the market. If you're the only or by far the largest customer, and have the power and weight to lean on just about all other potential customers, how free is the market?
Then the government changed its mind, stopped buying Crays, Convexes, MasPars, Connection Machines and Paragons, switched to CPU farms and IMPOSED PUNISHING TARIFFS ON SUPERCOMPUTER IMPORTS — thus forcing all its suppliers and grantees to switch to CPU farms as well. And this was a blatant act of meddling with the market too.
There never was a free market in the HPC business, so all this talk about how “free market failed us” or how “free market did just fine” is a fallacy.
As an HPC educator and practitioner though I know one thing. It was very easy to use the supercomputers of old and great many people, not necessarily the gurus and IT sophisticates, used them. But MPI clusters of today are so horrendously difficult to use that, even in research institutes, you will find only a handful of devotees who bother to write and run their own MPI programs. The most common use of these systems is capacity computing, which is not supercomputing at all. When used, occasionally, as a parallel computer resource, the clusters are energy and programmer-time inefficient, guzzling up mega-Watts of power, squandering most of it on empty cycles, and delivering little useful work in return.
Compare this whole HPC business, for example, to the car industry. Imagine that for some reason the government bans import of powerful engines and car manufacturers refuse to make engines or even special components for tractors and trucks. So now if you want to transport something really large and heavy, you have to assemble a cluster of PT Cruisers and pull your load this way.
This is more or less what we ended up doing in HPC.
Perhaps if the government *stopped* meddling in HPC and truly opened this market – including opening it to overseas competition – we would get our Connection Machines, Crays, MasPars, Paragons, Butterflies and Convexes back.
Regards,
Zdzislaw (Gustav) Meglicki,
Office of the Vice President for Information Technology,
Indiana University
HPCwire wants to know what YOU think. Please send comments to editor Tim Curns at [email protected].