On the opening page of his dissertation, which he submitted this past May, computer scientist George Porter cites a passage from the Hitchhiker’s Guide to the Galaxy series.
“The major difference between a thing that might go wrong and a thing that cannot possibly go wrong,” writes Douglas Adams in Mostly Harmless, “is that when a thing that cannot possibly go wrong goes wrong it usually turns out to be impossible to get at or repair.”
For Porter, who was part of a sketch comedy troupe when he was an undergraduate in the Department of Computer Science, the line is a way to sneak a smile into an otherwise dry technical document. It’s also about his research, which looks at how to diagnose problems that arise when Internet applications, like Google and Facebook, get so complex that they can’t be approached directly.
“Google just built a huge data center in Oregon. It’s a big warehouse with tens of thousands of linked computers,” says Porter, who’s now a research scientist at Sun Microsystems in San Diego. “Once you get to systems that are so large and so distributed, there’s simply too much information to deal with it as a whole. You have to treat the network like a statistical system.”
Porter’s main contribution to illuminating the interior of such “distributed” systems, which he developed over the course of his graduate study at University of California Berkeley, has been a program called “X-Trace.”
X-Trace allows a user to attach an additional set of instructions—a kind of digital hitchhiker—to what would otherwise be a typical request of an application. As the basic action (a web search, for instance) percolates through the application’s network, the hitchhiker keeps sending back information to the user about the environment through which it’s traveling.
The more of these journeys that are taken—and recorded—the more the accumulated “metadata” can be analyzed for patterns and anomalies. From there X-Trace is able to identify bugs, to notice where the application is working correctly, and to start building a meaningful model of what the network looks like.
Porter is continuing his work on X-Trace at Sun, which is one of the main funders, along with Google and Microsoft, of the lab he worked in at Berkeley. He’s also expanding into other questions that arise from the inherently complex nature of distributed computing systems.
“Right now, everyone is grappling with what’s the best way to reason about systems like this,” says Porter. “How do you build an operating system, or a physical machine, that can deal with these impossibly large amounts of data? I’m working with other people to figure out what the future should look like and to help facilitate that.”
This article also appeared in the Spring 2009 issue of Focus magazine.
Comments