P2P goes in search of intelligent applications

by George Lawton

February 15, 2001 —

Peer-to-peer (P2P) computing is generating a lot of excitement in some technical circles, but it is still a new technology looking for its niche. Much attention has been given to the P2P undertaking of industry luminary Ray Ozzie's Groove Networks; the company intends to provide a collaborative application environment that corporations may fancy. But another potential use of P2P lies in the realm of advanced distributed computing.

P2P distributed computing has commercial potential, but some of P2P's potential applications as a distributed platform put it more in line with recent trends in community-oriented open source software development. One of several precedents outlined for P2P distributed projects at this week's O'Reilly Peer-to-Peer Conference in San Francisco is the SETI@home project, which has set many technical professionals and hobbyists to the task of finding life elsewhere in the universe.

At a conference panel Wednesday, individuals discussed how the technology promises to rewire the backbone of the Internet. There is a whole spectrum of P2P applications from pure computation to pure data movement, said David Anderson, CTO of United Devices and director of the SETI@home distributed computing project. "The interesting stuff is somewhere in between," he said.

The SETI@home project exemplifies the possibilities at the first end of the P2P continuum. Over 2.7 million people run distributed computing applications in the background on their PC, which allows them to assist in the search for extraterrestrial intelligence. Together they process an average of 25 teraflops -- twice the capability of the second most powerful supercomputer on the planet. Despite the large number of machines, SETI@home is said to only require a staff of five to manage the entire system.

Andrew Chien, CTO of Entropia, one of the better-funded players in the distributed computing arena, speculates that at least 80 percent of all supercomputer applications could run across distributed networks of computers at a fraction of the cost of supercomputers. He explained, "I think there is a real opportunity here. The resources are quite cheap and underutilized. This type of technology will capture the largest fraction of the high-end market."

At the other end of the P2P continuum, companies are banking that they can develop more efficient and robust architectures than they can with traditional client/server systems. However, vendors in this arena will have to overcome IT fears about bringing unmanageable technology into the enterprise in order to succeed.

At this week's conference, Ozzie admitted that many IT managers were a bit leery of opening their enterprises' gates to Napster-like technologies that could overload their networks. He noted that P2P vendors have to approach IT management diplomatically, address their fears, and show how P2P technology can improve a company's capabilities.

Ozzie contrasted the approaches taken by Netscape Communications and PointCast, two potent players in the Web's early days. Netscape presented IT managers with a strategic vision of how the technology would grow to support the enterprise. In contrast, push-technology vendor PointCast essentially said its service was about their relationship with the customer. PointCast was consequently locked out of many enterprises.

Viewers suggest that of the dozens of companies jumping into the P2P arena, many have great technology but no apparent business model or source of revenue. Napster's legal and business difficulties were discussed extensively at the conference.

The appearance of so many cutting-edge technologists in search of a business model leads some to wonder if these early pioneers will go the way of some of the early players in pen computing or push technology.

Erick Von Schreiber, CTO of Cacheon, noted, "Peer to peer is already a successful technology." The issue, Von Schreiber said, is whether the companies now bringing it to market will be able to find a good business model.

ITworld.com