Reliability: Machine Learning May Make It Happen

When you think computers, "reliability" isn't the first word that comes to mind. (Can you say "blue screen"?)

Anyone who has lost data to a failed hard disk, power glitch, or similar heartbreak knows what I'm talking about. Which makes the goal of reliability that much more important.

Reliablity is just one of the goals of the Reliable, Adaptive and Distributed systems laboratory at the University of California, Berkeley. The RAD Lab, which launched in December 2005 under the direction of UC Berkeley professor of electrical engineering and computer sciences David Patterson, is adapting advances in statistical machine learning to the task of maintaining the large distributed computing systems needed to run data-intensive Internet businesses. These distributed systems typically require hundreds of engineers for development, debugging, and ongoing maintenance. The RAD Lab is designing software architecture to automate this.

The RAD Lab was launched with $7.5 million in funding over five years from Google, Microsoft, and Sun Microsystems. This week IBM, Hewlett-Packard, NTT Multimedia Communications Labs, Nortel Network, and Oracle signed on as affiliate members, each pledging annual contributions of up to $170,000 for the next five years.

All software and applications emerging from the RAD Lab will be made freely and openly available to the public, with source codes distributed using the Berkeley Software Distribution license. However, foundation and affiliate members of the RAD Lab will have access to the new technologies at least six months before the general public. Representatives from the companies will act as consultants and provide advice for the center's participants, but will not work at the RAD Lab.

Posted by Jon Erickson at 08:53 AM  Permalink