The Party Line Approach To Supercomputing
HAVE A MAJOR PROBLEM THAT demands a supercomputer, but limited funds? No problem, says University of Illinois at Chicago mathematician Robert Grossman. Build a "virtual supercomputer." The idea, adds Robert J. Hollebeek, a University of Pennsylvania collaborator, is to link a bunch of workstations "so they can `think'
Cluster computing isn't new. Sandia National Laboratories and Fermi National Accelerator Laboratory, among others, began working on cluster computing in the late 1980s. But coordinating the operations of multiple computers in different locations as they gang up on the same problem still required expensive hardware and software.
This team will rely on cheap off-the-shelf components and ATM circuit boards that plug into workstations. In this case, ATM is not a computerized bank teller--it means asynchronous transfer mode. This is the latest method for high-speed telephone-call switching. ATM is key, says Grossman, "because telephone signal quality doesn't degrade" as traffic on the network increases. The National Science Foundation is a believer. It plunked down $4 million to build a virtual supercomputer that will link 100 computers in three widely dispersed locations: University of Illinois' Chicago campus, University of Pennsylvania, and University of Maryland at College Park. If it works, other computers could be added, eventually creating the most powerful computer ever built.