4.3 Bringing Brains Together: Supercomputing and Grid Computing

As Moore’s Law makes possible the once impossible, businesses have begun to demand access to the world’s most powerful computing technology. SupercomputersComputers that are among the fastest of any in the world at the time of their introduction. are computers that are among the fastest of any in the world at the time of their introduction.A list of the current supercomputer performance champs can be found at http://www.top500.org. Supercomputing was once the domain of governments and high-end research labs, performing tasks such as simulating the explosion of nuclear devices, or analyzing large-scale weather and climate phenomena. But it turns out with a bit of tweaking, the algorithms used in this work are profoundly useful to business. Consider perhaps the world’s most well-known supercomputer, IBM’s Deep Blue, the machine that rather controversially beat chess champion Gary Kasparov. While there is not a burning need for chess-playing computers in the world’s corporations, it turns out that the computing algorithms to choose the best among multiple chess moves are similar to the math behind choosing the best combination of airline flights.

One of the first customers of Deep Blue technologies was United Airlines, which gained an ability to examine three hundred and fifty thousand flight path combinations for its scheduling systems—a figure well ahead of the previous limit of three thousand. Estimated savings through better yield management? Over fifty million dollars! Finance found uses, too. An early adopter was CIBC (the Canadian Imperial Bank of Commerce), one of the largest banks in North America. Each morning CIBC uses a supercomputer to run its portfolio through Monte Carlo simulations that aren’t all that different from the math used to simulate nuclear explosions. An early adopter of the technology, at the time of deployment, CIBC was the only bank that international regulators allowed to calculate its own capital needs rather than use boilerplate ratios. That cut capital on hand by hundreds of millions of dollars, a substantial percentage of the bank’s capital, saving millions a year in funding costs. Also noteworthy: the supercomputer-enabled, risk-savvy CIBC was relatively unscathed by the sub-prime crisis.

Modern supercomputing is typically done via a technique called massively parallelComputers designed with many microprocessors that work together, simultaneously, to solve problems. processing (computers designed with many microprocessors that work together, simultaneously, to solve problems). The fastest of these supercomputers are built using hundreds of microprocessors, all programmed to work in unison as one big brain. While supercomputers use special electronics and software to handle the massive load, the processors themselves are often of the off-the-shelf variety that you’d find in a typical PC. Virginia Tech created what at the time was the world’s third-fastest supercomputer by using chips from 1,100 Macintosh computers lashed together with off-the-shelf networking components. The total cost of the system was just $5.2 million, far less than the typical cost for such burly hardware. The Air Force recently issued a request-for-proposal to purchase 2,200 PlayStation 3 systems in hopes of crafting a supercheap, superpowerful machine using off-the-shelf parts.

Another technology, known as grid computingA type of computing that uses special software to enable several computers to work together on a common problem as if they were a massively parallel supercomputer., is further transforming the economics of supercomputing. With grid computing, firms place special software on its existing PCs or servers that enables these computers to work together on a common problem. Large organizations may have thousands of PCs, but they’re not necessarily being used all the time, or at full capacity. With grid software installed on them, these idle devices can be marshaled to attack portions of a complex task as if they collectively were one massively parallel supercomputer. This technique radically changes the economics of high-performance computing. BusinessWeek reports that while a middle-of-the-road supercomputer could run as much as thirty million dollars, grid computing software and services to perform comparable tasks can cost as little as twenty-five thousand dollars, assuming an organization already has PCs and servers in place.

An early pioneer in grid computing is the biotech firm Monsanto. Monsanto enlists computers to explore ways to manipulate genes to create crop strains that are resistant to cold, drought, bugs, pesticides, or that are more nutritious. Previously with even the largest computer Monsanto had in-house, gene analysis was taking six weeks and the firm was able to analyze only ten to fifty genes a year. But by leveraging grid computing, Monsanto has reduced gene analysis to less than a day. The fiftyfold time savings now lets the firm consider thousands of genetic combinations in a year.Peter Schwartz, Chris Taylor, and Rita Koselka, “The Future of Computing: Quantum Leap,” Fortune, August 2, 2006. Lower R&D time means faster time to market—critical to both the firm and its customers.

Grids are now everywhere. Movie studios use them to create special effects and animated films. Proctor & Gamble has used grids to redesign the manufacturing process for Pringles potato chips. GM and Ford use grids to simulate crash tests, saving millions in junked cars and speeding time to market. Pratt and Whitney test aircraft engine designs on a grid. And biotech firms including Aventis, GlaxoSmithKlein, and Pfizer push their research through a quicker pipeline by harnessing grid power. JP Morgan Chase even launched a grid effort that mimics CIBC’s supercomputer, but at a fraction of the latter’s cost. By the second year of operation, the JPMorgan Chase grid was saving the firm five million dollars per year.

Massive clusters of computers running software that allows them to operate as a unified service also enable new service-based computing models, such as software as a service (SaaS)A form of cloud computing where a firm subscribes to a third-party software and receives a service that is delivered online. and cloud computingReplacing computing resources—either an organization’s or individual’s hardware or software—with services provided over the Internet.. In these models, organizations replace traditional software and hardware that they would run in-house with services that are delivered online. Google, Microsoft, SalesForce.com, and Amazon are among the firms that have sunk billions into these Moore’s Law–enabled server farmsA massive network of computer servers running software to coordinate their collective use. Server farms provide the infrastructure backbone to SaaS and hardware cloud efforts, as well as many large-scale Internet services., creating entirely new businesses that promise to radically redraw the software and hardware landscape while bringing the gargantuan computing power to the little guy (see Chapter 10 "Software in Flux: Partly Cloudy and Sometimes Free").

You can join a grid, too. SETI@Home turns your computer screen saver into a method to help “search for extraterrestrial intelligence,” analyzing data from the Arecibo radio telescope system in Puerto Rico (no E.T. spotted yet). FightAids@Home will enlist your PC to explore AIDS treatments. And Folding@Home is an effort by Stanford researchers to understanding the science of protein-folding within diseases such as Alzheimer’s, cancer, and cystic fibrosis. A version of Folding@Home software for the PlayStation 3 had enlisted over half a million consoles within months of release. Having access to these free resources is an enormous advantage for researchers. Says the director of Folding@Home, “Even if we were given all of the NSF supercomputing centers combined for a couple of months, that is still fewer resources than we have now.”G. Johnson, “Supercomputing ‘@Home’ Is Paying Off,” New York Times, April 23, 2002.

Multicore, massively parallel, and grid computing are all related in that each attempts to lash together multiple computing devices so that they can work together to solve problems. Think of multicore chips as having several processors in a single chip. Think of massively parallel supercomputers as having several chips in one computer, and think of grid computing as using existing computers to work together on a single task (essentially a computer made up of multiple computers). While these technologies offer great promise, they’re all subject to the same limitation: software must be written to divide existing problems into smaller pieces that can be handled by each core, processor, or computer, respectively. Some problems, such as simulations, are easy to split up, but for problems that are linear (where, say, step two can’t be completed until the results from step one are known), the multiple-brain approach doesn’t offer much help.

Moore’s Law will likely hit its physical limit in your lifetime, but no one really knows if this “Moore’s Wall” is a decade away or more. What lies ahead is anyone’s guess. Some technologies, such as still-experimental quantum computing, could make computers that are more powerful than all the world’s conventional computers combined. Think strategically—new waves of innovation might soon be shouting “surf’s up!”