For years, multitudes of users have worked confidently on their computers, knowing that a nearby server is storing files, providing compute cycles and providing them with the necessary maintenance. Professor Peter Druschel wants to take all of that away ... and give much more in return.
Druschel, a leader in the Systems group, is jointly leading a five-year, $12 million, multi-institutional program to develop a cooperative or peer-to-peer framework that will support the deployment of large, resilient distributed applications and reduce the need for servers. Druschel's project, entitled IRIS, is funded through the National Science Foundation's Information Technology program.
"Today, everyone relies on dedicated hardware and software for services like file storage, backup, email and Web content distribution," Druschel said. "It's quite expensive, requires a lot of system administration, and is vulnerable to failures and security attacks. We're interested in creating an ecology where people can contribute some of their resources for the common good, and get highly available, robust services in return that they could not otherwise afford.
Most users utilize a small percentage of their computer's capacity. Pastry would use each computer as a node in a self-organizing overlay network, thus decentralizing the system. The resulting network would be cheaper, reducing the need for servers, machine rooms, and system administration. Additionally, there would be no single point of failure, and no single points of attack.
"Small operators can't afford highly available systems, but can pool resources of their existing systems to provide service, and give and take," Druschel explained. "They can help each other by forming a joint infrastructure that is highly robust and scales with the size of the community." Disk back up, which is quite costly, would be unnecessary since the software would encrypt the data and parcel it out for storage.
The project involves a dozen teams at several universities and research institutes, including the University of California, Berkeley, New York University, the Massachusetts Institute of Technology and the ICSI Center for Internet Research.
December 18, 2002