Supercomputer built with 1,100 PCs

John Markoff

A supercomputer is a computer that leads the world in terms of processing capacity, particularly speed of calculation. The term "super computing" was first used by New York World newspaper in 1920 to refer to large custom-built tabulators IBM made for Columbia University. Supercomputers introduced in the 1960s were designed primarily by Seymour Cray at Control Data Corporation (CDC), and led the market into the 1970s until Cray left to form his own company, Cray Research. He then took over the supercomputer market with his new designs, holding the top spot in supercomputing for 5 years (1985-1990).
   Cray himself never used the word "supercomputer," a little-remembered fact in that he only recognised the word "computer." In the 1980s a large number of smaller competitors entered the market, in a parallel to the creation of the minicomputer market a decade earlier, but many of these disappeared in the mid-1990s "supercomputer market crash". Today, supercomputers are typically one-of-a-kind custom designs produced by "traditional" companies such as IBM and HP, who had purchased many of the 1980s companies to gain their experience, although Cray Inc. still specializes in building supercomputers.
   The Cray-2 was the world's fastest computer from 1985 to 1989.The term supercomputer itself is rather fluid, and today's supercomputer tends to become tomorrow's normal computer. CDC's early machines were simply very fast scalar processors, some ten times the speed of the fastest machines offered by other companies. In the 1970s most supercomputers were dedicated to running a vector processor, and many of the newer players developed their own such processors at a lower price to enter the market. The early and mid-1980s saw machines with a modest number of vector processors working in parallel become the standard. Typical numbers of processors were in the range 4-16. In the later 1980s and 1990s, attention turned from vector processors to massive parallel processing systems with thousands of "ordinary" CPUs, some being off the shelf units and others being custom designs. (This is commonly and humorously referred to as the attack of the killer micros in the industry.) Today, parallel designs are based on "off the shelf" server-class microprocessors, such as the PowerPC, IA-64, or x86-64, and most modern supercomputers are now highly-tuned computer clusters using commodity processors combined with custom interconnects.
   Software tools
   Software tools for distributed processing include standard APIs such as MPI and PVM, and open source-based software solutions such as Beowulf and openMosix which facilitate the creation of a sort of "virtual supercomputer" from a collection of ordinary workstations or servers. Technology like ZeroConf (Rendezvous/Bonjour) pave the way for the creation of ad hoc computer clusters. An example of this is the distributed rendering function in Apple's Shake compositing application. Computers running the Shake software merely need to be in proximity to each other, in networking terms, to automatically discover and use each other's resources. While no one has yet built an ad hoc computer cluster that rivals even yesteryear's supercomputers, the line between desktop, or even laptop, and supercomputer is beginning to blur, and is likely to continue to blur as built-in support for parallelism and distributed processing increases in mainstream desktop operating systems. An easy programming language for supercomputers remains an open research topic in Computer Science.
   Uses
   Supercomputers are used for highly calculation-intensive tasks such as problems involving quantum mechanical physics, weather forecasting, climate research (including research into global warming), molecular modeling (computing the structures and properties of chemical compounds, biological macromolecules, polymers, and crystals), physical simulations (such as simulation of airplanes in wind tunnels, simulation of the detonation of nuclear weapons, and research into nuclear fusion), cryptanalysis, and the like. Major universities, military agencies and scientific research laboratories are heavy users.
   A particular class of problems, known as Grand Challenge problems, are problems whose full solution require semi-infinite computing resources. A home-brew supercomputer, assembled from off-the-shelf personal computers in just one month at a cost of slightly more than $5 million, was about to be ranked as one of the fastest machines in the world a couple of years ago.
   Word of the low-cost supercomputer, put together by faculty, technicians and students at Virginia Polytechnic Institute, is shaking up the esoteric world of high performance computing, where the fastest machines have traditionally cost from $100 million to $250 million and taken several years to build.
   The Virginia Tech supercomputer, put together from 1,100 Apple Macintosh computers, has been successfully tested in recent days, according to Jack Dongarra, a University of Tennessee computer scientist who maintains a listing of the world's 500 fastest machines.
   The official results for the ranking will not be reported until next month at a supercomputer industry event. But the Apple-based supercomputer, which is powered by 2,200 I.B.M. microprocessors, was able to compute at 7.41 trillion operations a second, a speed surpassed by only three other ultra-fast computers.
   The fastest computers on the current Top 500 list are the Japanese Earth Simulator; a Los Alamos National Laboratory machine dedicated to weapons design; and another weapons oriented cluster of Intel Pentium 4 microprocessors at the Lawrence Livermore National Laboratories. Officials at the school said that they were still finalizing their results and that the final speed number might be significantly higher.
   "We are demonstrating that you can build a very high performance machine for a fifth to a tenth of the cost of what supercomputers now cost," said Hassan Aref, the dean of the School of Engineering at Virginia Tech in Blacksburg, Va. The computer was put together in a virtual flash. Scientists from the school met with Apple executives two days after the company introduced its new 64-bit desktop computer.
   Apple agreed to put the school at the head of the line for the new machines. Starting when they returned to school in September, student volunteers, who received free pizzas for their labour, helped with the assembly of the system, essentially an array of large refrigerators to keep the computers from overheating. Virginia Tech's president offered free football tickets to the technicians who were spending long hours on the project.
   The ranking is a coup for Apple, which for several years has lagged behind, in terms of raw computing speed, the PC world controlled by Intel and Advanced Micro Devices microprocessors. It is also an indication that the supercomputer industry, which has been in eclipse since the end of the cold war, is again playing a more vital role.
   "On the surface this is a pretty impressive machine," Mr. Dongarra said. "It shows that the processors are getting to the point where this kind of performance will be quite common."
   The performance of the new computer highlights the challenge to highly expensive custom-designed machines - like the Earth Simulator of Japan, which is assembled from 5,120 custom processors that have special circuitry for performing long strings of mathematical operations - from computers put together by linking more common off-the-shelf components in fairly simple ways.
   The Japanese computer was measured at 35.8 trillion operations a second last year but American computer experts estimate that it cost as much as $250 million. By contrast, the fastest cluster machine, the Lawrence Livermore system consisting of 2304 Intel Xeon processors, is capable of 7.63 trillion operations a second, at a price estimated at $10 million to $15 million. The Virginia Tech computer makes the cost-to-performance equation even starker.
   Japanese Computenik beats US supercomputer
   The Japanese Earth Simulator, writes Ad Emmen, is on-line and producing results that alarm the USA, that considered itself as being leading in supercomputing technology. With over 35 Tflop/s, it five times outperforms the Asci White supercomputer that is leading the current TOP500 list. No doubt that position is for the Earth Simulator, not only for the next list, but probably even for the coming two years. In the New York Times, bench mark compiler Jack Dongarra compares the event with the Sputnik, hence he dubbed the Earth Simulator "Computenik".
   The Earth Simulator is a special purpose machine, made by NEC with the same type of vector technology as is available on the SX-6. This is a different approach than the Americans are taking, who base the large machines on off-the shelf technology. A clear advantage of the Japanese approach is that they "only" need 5104 processors for the Earth Simulator. The Asci White at the US Lawrence Livermore National Laboratory and made by IBM, already needs 8192 processors for 7 Tflop/s.
   Nevertheless, a complete new building was created for the Earth Simulator, because it is a huge machine.
   The Earth Simulator will be used for research in Earth related science. It is especially suited for simulating complex linked systems, where for instance, the climate is modelled together with water flow on the earth and models of the ocean. In general, these types of applications lend themselves very good for vector processing.
   Hans Meuer, the father of the TOP500 list of the world's most powerful supercomputers, asked for a comment, says: The ES (Earth Simulator) is a significant step into the future of high end supercomputers: it is approximately 5 times faster than the current #1, ASCI White, if we take the best linpack performance as a yardstick. I do expect the ES to be faster than the sum of the first other 19 machines in the TOP20 of the forthcoming 19th TOP500 list. Thus the ES is a real challenge for the US ASCI program. The US labs are now falling significantly behind.
   The New York Times reports: "In some sense we have a Computenik on our hands," said Jack Dongarra, a University of Tennessee computer scientist who reported the achievement today. On this citation, Hans Meuer comments: The exciting thing for me is that this is the first system after Numerical Wind Tunnel (built by Fujitsu) to take over the lead in the TOP500 list as a true vector processing system. And like the Numerical Wind Tunnel being the number one spot from 11/93 to 11/95 I expect the ES to be the number one in the world for at least two years.
   In Europe we do not have a supercomputer industry, we do have, however many supercomputer applications. What will be the impact on "Earth simulation" type of research in Europe? We are pretty good at that today. Are the large supercomputers we have here (and the Grid based projects) powerful enough to keep up with the Japanese research in this field? We did ask Hans Meuer, who said: It is difficult to judge on the impact of ES in Europe. But I think that vector processing with a very long tradition especially in Europe could experience a renaissance as one of the leading architectures. In Europe there is no major HPC manufacturing activity at present, therefore we should appreciate the diversity of different architectures of HPC. NEC did a great job with the ES as a major step to emphasize that vector processing is not only still alive but capable to take the number one spot in the world with an unbelievable gap to the other high end systems. "Earth Simulator" type of research in Europe has already decided, at least partially, to run their problems on SX-6 based systems, e.g. the "Deutsche Klimarechenzentrum (DKRZ)" in Hamburg.

^ TOP OF THIS PAGE ^ MAIN PAGE

Brad Stone and Miguel Helft

As media companies struggle to reclaim control over their movies, television shows and music in a world of online file-sharing software, they have found an ally in software of another kind.
   The new technological weapon is content-recognition software, which makes it possible to identify copyrighted material, even, for example, from blurry video clips.
   The technology could address what the entertainment industry sees as one of its biggest problems - songs and videos being posted on the Web without permission.
   Vance Ikezoye, the chief executive of Audible Magic in Los Gatos, Calif., demonstrated the technology by downloading a two-minute clip from YouTube and feeding it into his company's new video-recognition system.
   The clip - drained of color, with dialogue dubbed in Chinese - appeared to have been recorded with a camcorder in a dark movie theater before it was uploaded to the Web, so the image quality was poor.
   Still, Mr. Ikezoye's filtering software quickly identified it as the sword-training scene that begins 49 minutes and 37 seconds into the Miramax film "Kill Bill: Vol. 2."
   The entertainment industry is clamoring for Internet companies to adopt the technology for music files as well as for video clips. The social networking site MySpace, owned by the News Corporation, said last week that it would use Audible Magic's system to identify copyrighted material on its pages. But not every Internet company is rushing to go along. The video-sharing site YouTube, which Google bought last year, is the major holdout so far.
   Though YouTube's co-founders said publicly that they would start using filtering technology by the end of last year, the site has yet to do so. And they have further angered some media companies by saying they would only use such technology to detect clips owned by companies that agree to broader licensing deals with YouTube.
   The pressure is on. Executives at media companies like NBC and Viacom have criticized Google for the delay. Earlier this month, Viacom asked YouTube to remove 100,000 clips of its shows, like music videos from MTV and excerpts from Comedy Central's "The Daily Show."
   In a statement, YouTube said that identifying which video clips had been uploaded without permission was a complex problem that required the cooperation of the copyright owners. "On YouTube, identifying copyrighted material cannot be a single automated process," it said in the statement.
   The systems being developed by companies like Audible Magic and Gracenote make use of vast databases that store digital representations of copyrighted songs, TV shows and movies.
   When new files are uploaded to a Web site that is using the system, it checks the database for matches using a technique known as digital fingerprinting. Copyrighted material can then be blocked or posted, depending on whether it is licensed for use on the site.
   "This is capable of helping the film and TV studios comprehensively protect their works," Mr. Ikezoye said. "This could put the genie back in the bottle."
   Audio fingerprinting technologies have been used successfully for some time to detect copyrighted music on file-sharing networks and, to a smaller degree, to identify music tracks on social-networking Web sites like MySpace.
   Systems that can identify video files hold even greater promise to improve relations between traditional media companies and Internet companies like YouTube. But the technology is not quite ready.
   "Video is much more complex to analyze, and more information needs to be captured in the fingerprint," said Bill Rosenblatt, president of GiantSteps Media Technology Strategies, a consulting firm based in New York. He noted that there were also more ways to fool the technology - for example, by cropping the image.
   Screening for video is also more difficult because of the sheer volume of new material broadcast on television each day, all of which must be captured in the database.
   And deploying any type of fingerprinting technology can carry a price. Users tend to leave filtered Web sites and migrate to more anything-goes online destinations.
   Nevertheless, some file-sharing networks and smaller video sites like Guba.com and Grouper.com are already using more basic filters that monitor video soundtracks and music files, hoping to appease copyright holders and stay out of the courtroom.
   Last week, they got some company: MySpace announced that it would expand on early filtering efforts and license Audible Magic's audio and video fingerprinting technology. It will use the system to identify and obtain authorization for material from Universal Music, NBC Universal and Fox, three media companies that have wanted more control over their content on the site. The move ratchets up the pressure on YouTube, the largest video site on the Web.
   A coalition of record companies sued the operators of the file-sharing program LimeWire for copyright infringement on Friday, claiming the company encouraged users to trade music without permission. The Recording Industry Association of America said in a statement that it had sued the Lime Group, the corporation's executives.
   Hollywood, long tormented by digital piracy, is growing excited about the possibilities of digital fingerprinting and filtering - in part because it is tired of having to ask YouTube and other sites to remove individual clips, only to find them posted again by other users.
   "To the extent you can readily and easily identify one film or TV show from the next, it enables different licensing models and the opportunity to protect your content," said Dean Garfield, executive vice president of the Motion Picture Association of America.
   For now, however, audio fingerprinting is all that is widely available, and it can fall short in some situations, like when someone pairs a song with an unrelated piece of video.
   For example, last December, one YouTube user uploaded scenes from the Warner Brothers movie "Superman Returns," matched to the song "Superman," by Five for Fighting of Columbia Records, a unit of Sony BMG Music.
   With acoustic fingerprinting, Sony could authorize the use of the song and get a slice of the advertising revenue the clip generates, but Warner Brothers could not because the filter does not scrutinize video images.
   Hoping to nurture the development of more advanced video fingerprinting, the film association asked technology companies last fall to submit video filtering systems for testing. Mr. Garfield of the association said 13 companies responded; their systems are now being evaluated.
   Perhaps not surprisingly, there is now a flurry of interest in digital fingerprinting in Silicon Valley. Sean Varah, an electronic-music researcher who once worked for Sony music's venture capital group, founded the start-up MotionDSP in 2005 to develop technology to improve the quality of video images. But he changed the company's direction last year after seeing an opportunity in the filtering business.
   "The television and movie producers have learned a lesson from Napster," he said, referring to the music-sharing service that first got the attention of media companies. "They are not going to wait and see what happens."
   Attributor, another start-up based in Redwood City, Calif., is taking a different approach to filtering. It is developing automated software that will travel the Internet looking for copyrighted text, audio and video.
   Setting up filters for each and every Web site and peer-to-peer network "is not a long-term solution," said Jim Brock, a former Yahoo executive and the chief executive of Attributor. Rights holders "need to have these kinds of solutions across the Internet," he said.
   Audible Magic, which is considered to be an early leader in the field, started out with a system to recognize songs played on the radio, so it could offer listeners an opportunity to buy the music online. The company later adapted that technology to create an audio fingerprinting system.
   Mr. Ikezoye, a former Hewlett-Packard marketing executive, recently set out to expand into video recognition. Last year, he licensed an invention called Motional Media ID, created by David W. Stebbings, a former executive at the Recording Industry Association of America.
   Neither Mr. Ikezoye nor Mr. Stebbings would offer details on Motional Media ID (which identified the "Kill Bill" clip), citing the newly competitive environment around digital fingerprinting. Mr. Ikezoye acknowledged that it did not work well for very short clips and said that he would probably have to buy or develop additional technology.
   Deploying any type of fingerprinting filter can have both good and bad effects. Guba.com, a video-sharing site similar to YouTube, developed its own filtering system, which it calls Johnny. Having won the favor of the film industry, the company now has deals to sell Warner and Sony films on its site.
   But when Guba began blocking many copyrighted clips last April, its popularity plunged.
   "We took a huge hit," said Eric Lambrecht, Guba's chief technology officer. "We all know what people want to see, but we looked at it as a long-term business decision."
   Some experts believe wide adoption of the technology is inevitable.
   "As technology companies mature, they are realizing that the rule of law is better than the anarchy in which they were formed," said Paul Kocher, chief executive of Cryptography Research, a company that has studied the security of digital fingerprinting technology.
   Courtesy: N Y Times

^ TOP OF THIS PAGE ^ MAIN PAGE