Friday, August 25, 2000

A New Protocol Helps Researchers Share Large Data Sets Over Fast Networks By FLORENCE OLSEN Researchers at the University of Illinois at Chicago say they have moved a step closer to perfecting an Internet protocol that will make it easier for scientists to share their data over super-fast research networks. Documents of all kinds are now easily shared over the World Wide Web, facilitated by the hypertext-transfer protocol, or http. But even scientists using the fastest research networks still have no easy way to exchange rows and columns of raw data with other researchers in real time, says Robert L. Grossman, the director of the university's Laboratory for Advanced Computing. A scientist who studies climate data, for example, might be interested in whether there is a real correlation between sunspot activity and changes in climate. Using the free client-and-server software developed at Illinois, climate scientists could, "with a few clicks, align the data sets by year to see if there might be something there," Mr. Grossman says. "At that point, they could do a more in-depth analysis of the data." Mr. Grossman says the difficulty of using distant and scattered online data archives has inhibited scientists from mining many sources that might improve the accuracy of climate models and other computationally intensive forecasting tools. He and his colleagues have devised what they say is a way to move large sets of data over the next generation of online networks quickly and inexpensively. Their method relies on, among other things, a new protocol that they call the data-space-transfer protocol. It moves move huge amounts of data over research networks at speeds fast enough to sustain a computation on clusters of Linux and Windows NT computers at the other end, Mr. Grossman says. The transfer protocol, which works in tandem with other software developed at Illinois, accomplishes the task by partitioning the data and sending it across several open network connections simultaneously. The new protocol lets scientists publish their data online in a simple format known as a flat file, which saves researchers the bother of creating complex relational databases, says Ted Hanss, director of applications development for the Internet2 consortium. The protocol also makes it easy for scientists to tap into other research data at distant locations, he says. "It doesn't require the data to be located all in one place." Last week the Illinois researchers, together with other institutions and government agencies participating in a terabyte challenge testbed, set a new speed record of 250 megabits per second when they used the protocol to sustain a computation that ran for three days over several high-speed network-backbone connections. The high-speed-transfer protocol moved bulky climate data from 21 servers at three academic-research sites to servers at the National Aeronautics and Space Administration's Ames Research Center in California. Researchers can download free client and server versions of the Java software and source code at http://www.dataspaceweb.net Mr. Grossman says he hopes the data-transfer protocol becomes a standard that is widely used. "What we're trying to do is make it simple for scientists to put data out there on the Web." Easy-to-print version E-mail this article

George W. Bush urges aid for minority institutions; Al Gore seeks to make college affordable Girls catch up with boys in math, report finds, but overall student gains have slowed A new protocol helps researchers share large data sets over fast networks Measuring students' learning is a major challenge for distance education