Boffins devise P2P supercomputer

A team of scientists has worked out a way of making peer-to-peer (P2P) networks run at the speed and efficiency of supercomputers.

A team from Princeton University in New Jersey, the University of California in Berkley and the networking companies AT&T and Cisco have developed a system of "random walkers" to speed up the process, according to New Scientist magazine.

Rather than flooding many messages across a network, the system has discovered that it is better to allow a few messages to randomly "walk" between individual machines.

The research indicated that the optimum number of "walkers" sent out to find a file is between 16 and 64.

The group hopes its research can be used to distribute massive amounts of computing power for scientific applications in the form of "distributed supercomputers".

But it could also boost internet music and video file sharing by making the underlying P2P networks larger and more robust.

One of the researchers, Princeton's Qin Lv, told New Scientist: "The current scaling problems that Gnutella is having is mainly due to its flooding search algorithm, which generates too much network traffic.

"Any search algorithm that can effectively control its message traffic would be beneficial in terms of scalability."