Solar's Big Future with Small Tech

Electronic Nose that Knows

Cool Alumni

Dean's Digest

Printer-friendly version

During his long career, professor David Patterson led the design and implementation of Reduced Instruction Set Computer (RISC) and, with professor Randy Katz, Redundant Arrays of Inexpensive Disks (RAID) technology, both of which became industry standards.

In October, five robotic vehicles drove themselves more than 130 miles across the Mojave Desert, making international headlines and marking a milestone in machine learning. Related to artificial intelligence, statistical machine learning refers to methods that enable a computer to improve its performance by analyzing previous results. Now, a group of UC Berkeley researchers hope to bring the same technology into cyberspace. The new Reliable, Adaptive and Distributed systems Laboratory (RAD Lab), funded with $7.5 million from Google, Microsoft, and Sun Microsystems, is developing technology that leverages the power of statistical machine learning so that one person can create the next eBay, Amazon, or even Google, by him or herself.

"eBay has changed the world but they had to hire hundreds of really smart people to pull it off," says computer science professor David Patterson, founding director of the RAD Lab. "Wouldn't it be great if one person could create a eBay-sized system without an eBay-sized organization?"

To make this vision a reality, the RAD Lab, which falls under the Center for Information Technology Research in the Interest of Society (CITRIS), must weave together innovations in such diverse disciplines as networking, computer architecture, systems theory, and statistics. To that end, Patterson is collaborating with a handful of UC Berkeley professors who are recognized as pioneers in those areas: Randy Katz, Scott Shenker, Ion Stoica, Armando Fox (joining Berkeley next summer), and Michael Jordan, who holds a joint position in computer science and the Department of Statistics.

In an early experiment, the RAD Lab researchers created "operator-friendly" visualizations to cross check results generated by machine learning algorithms. Analyzing real outage data provided by Ebates.com, the researchers determined that many failures could have been predicted hours in advance. (courtesy RAD Lab) [View larger image]

The RAD Lab calls for a new paradigm in the traditional software design and development process. The current model of software design follows a step-by-step path from the initial idea through development, testing, deployment, to ongoing operation. According to Patterson though, that "waterfall" model is obsolete. It's far too slow and hierarchical for Internet applications where the software behind Web services is constantly being tweaked, improved, and scaled up to accommodate growing numbers of users.

The RAD Lab embraces a very different approach where the operation of the Web applications and services is measured in real-time. Insights gleaned from that data is immediately fed back into the development process to improve the code. The model, Patterson explains, "shortens the 'distance' between a service's users and its developers and allows for faster innovation and bug fixing." And therein lies the rub.

"How can a single human keep track of everything, fix problems quickly, and make sure that the service never goes off the air?" Patterson says.

The answer, he says, is to automatically close the feedback loops using statistical machine learning techniques. Statistical machine learning employs pattern-finding algorithms that compare historical data and find commonalities between them. Those commonalities can then be used to generate a predictive model of what's likely to happen in the future.

"The strength of machine learning is sifting through massive amounts of data and finding insights into what's going on," Patterson says. "So we're trying to come up with statistical methods to measure and observe what's happening within Web services and notice indications of problems far enough in advance that problems can be prevented. Then, techniques taken control theory could be used to automatically help fix the things that are wrong."

Do you have a comment or question regarding this research? We want to hear from you...

Initially, RAD Lab research will be conducted by the faculty co-founders and ten computer science graduate students. The number of collaborators is expected to grow as the research progresses. From day one though, all of the software developed within the RAD Lab will be available to the public through the Berkeley Software Distribution license, a revolutionary "open source" approach created at the University in the 1970s that enables anyone to build upon, change, and use the raw programming code free of charge.

"Our vision is to enable one person to invent and run the next revolutionary Internet service, operationally expressing a new business idea as a multi-million-user service over the course of a long weekend," Patterson says.

David Patterson's home page

Randy Katz's home page

Michael Jordan's home page

Ion Stoica's home page

Armando Fox's home page

Scott Shenker's home page

"Seeing Patterns" by David Pescovitz (Lab Notes, May 2004)

"Berkeley UNIX and the Birth of Open-Source Software" (Lab Notes, October 2001)

Lab Notes is published online by the Marketing and Communications Office of the UC Berkeley College of Engineering. The Lab Notes mission is to illuminate groundbreaking research underway today at the College of Engineering that will dramatically change our lives tomorrow.

Media contact: Teresa Moore, Lab Notes editor, Director of Marketing and Communications
Writer, Researcher: David Pescovitz
Web Manager: Michele Foley

Subscribe or send comments to the Engineering Marketing and Communications Office: lab-notes@coe.berkeley.edu.

© 2006 UC Regents. Updated 1/2/06.