The University of California at Berkeley wants to create a data storage network that encompasses the planet.
OceanStore is a research project at the university that would use software to break data into many tiny, encrypted parts and store them across a vast array of Web servers owned by Internet service providers around the world.
A vast redundant storage network such as OceanStore would afford easy access to data from anywhere and unprecedented levels of disaster recovery, according to its inventor, John Kubiatowicz. If more than one computer or server were to crash, OceanStore would be able to rebuild the information using pieces stored in multiple clusters on other servers.
OceanStore would track documents by assigning each one a globally unique identification (GUID) tag before it’s split into fragments and sent over the Internet to be stored randomly throughout the network.
“You would maybe spread 64 fragments of a document around, and maybe 16 of those can be used to reconstruct [the original document],” Kubiatowicz said. “We’re assuming a system the scale of OceanStore will have pieces of it broken all the time.”
For example, to retrieve a chopped-up 1989 tax return, OceanStore would send intelligent agents onto the Internet looking for a GUID tag, he said. As the messenger agents search, they would leave behind trails of digital bread crumbs so that the next time, the agents could find the data more quickly.
More frequently used data would be stored on nearby servers to cut down on latency.
Consumers who want to save their documents on OceanStore would pay a monthly fee to an Internet service provider, which would then arrange to redundantly store the data on another Internet provider’s server for a small fee.
Neal Goldman, a research analyst at The Yankee Group Inc. in Boston, said there are inherent problems associated with OceanStore. For example, enterprises probably wouldn’t store mission-critical data on a Web-based system.
“You don’t know what the bandwidth is between you and the piece of data you want, so there are some real issues as far as performance,” Goldman said.
The project has received about US$500,000 in seed funding from vendors including IBM Corp., Nortel Networks Corp. in Brampton, Ont., EMC Corp. in Hopkinton, Mass., and federal agencies such as the National Science Foundation and the Defense Advanced Research Projects Agency, both in Arlington, Va., Kubiatowicz said.
