| DISTRIBUTED COMPUTING: POWER GRID By Marianne Sullivan In contrast to modern networks, a lot of older, centralized systems were akin to an elephant -- huge and powerful, yes, but very difficult to replace, he says. Zhou, who started using these kinds of analogies while earning his Ph.D. in computer science at the University of California, Berkeley, in the mid-1980s, co-founded Platform in 1992 to see if he could give them a practical application. His idea: that generic, networked computers should be easy to configure and, when they’d outlived their usefulness, easy to dispose of. But the point wasn’t just to create a more reliable system; Zhou also wanted to offer a way to farm out the most complex number-crunching tasks to hundreds or even thousands of clustered computers acting in concert. Zhou’s brand of distributed computing had immediate appeal among engineering-intensive industrial companies like Boeing Co. and Motorola, as well as government research labs and their academic collaborators working on joint projects. The concept later found a more popular use in Napster, which hooked millions of personal computers into a shared virtual music library. Privately held Platform has grown to serve 1,500 customers utilizing the technology -- now known as grid computing -- for purposes ranging from genetic research to film animation. The financial industry has awakened to grid computing over the past two years. Platform’s biggest client, J.P. Morgan Chase & Co., has introduced it into risk management and derivatives-pricing applications, and other projects are on the drawing board, says Steven Neiman, the bank’s vice pres- ident of high-performance computing. Vendors estimate that 20 percent of Wall Street firms have adopted the technology. Along with two other computing trends -- the open-standard Linux operating system and Web services -- grid computing offers timely solutions for technologists on tight budgets. Major financial firms typically maintain thousands of desktop and server computers that rarely operate at more than 20 percent of capacity. Grid computing harnesses and redistributes that unused power. To run, say, a complex multivariate financial model, grid software draws the necessary power from underutilized machines. Grids are “fungible pools of processing and storage [that can be] allocated on demand,” explains Vern Brownell, a former Goldman, Sachs & Co. chief technology officer who is founder and CTO of Marlboro, Massachusettsbased software company EGenera. Corporations save money in two ways: They have less need to buy new hardware; and they can run more programs, since analyses that might have taken hours to run on a single computer can now be completed in minutes. “It is a way to combine cheap chips with great interconnectability and data mobility to get easy access to vast, cheap CPU [central processing unit] power,” says J.P. Morgan’s Neiman. The financial markets have spawned their own specialists in grid architecture and in a variant called server virtualization, which increases efficiency by turning a single computer into several “virtual machines.” These suppliers include Brownell’s EGenera, which provides trading technology to Credit Suisse First Boston; and DataSynapse, a New York outfit that has sold its system to Charlotte, North Carolinabased Wachovia Securities and to Abbey National in the U.K. Merrill Lynch & Co. is implementing server virtualization. Working with Palo Alto, Californiabased VMware, Merrill plans over the next two to three years to convert thousands of computers into what Merrill CTO John McKinley calls a “logical mainframe.” Server virtualization differs from the ordinary grid approach, says McKinley, because it’s not necessarily geared for computation-intensive operations and doesn’t require reengineering of existing applications. However, the objective -- “maximum bang for the buck” -- is the same, adds McKinley. “In terms of software licensing, labor and hardware costs, we’ll save millions of dollars a year.” |
| ©Copyright 2003 Institutional Investor, Inc. |
