The Cavalier Daily
Serving the University Community Since 1890

Legion of computers create supernetwork

A computer program created at the University is changing the way technology experts think about computer networking.

University researchers led by Associate Computer Science Professor Andrew Grimshaw invented Legion, a software system that acts like a sandwich between a normal PC and other PCs on a network to create a giant virtual supercomputer.

"Legion is in a good position now," said Associate Computer Science Prof. Marty Humphrey, a member of the research team. One of a new class of distributive computing programs, Legion and software like it were the focus of a recent conference sponsored by Intel -- a computer processor manufacturer.

A distributive computer program breaks down a massive problem into smaller tasks, which is then solved by individual PCs. The most famous distributive program is "SETI@home," a screen saver available for download on the World Wide Web that uses the computing power of hundreds of thousands of PCs to process radio signals from space.

Distributive computing software signifies a paradigm shift in the field of networking and data processing. In the past, researchers relied on centralized systems controlled by one or a few supercomputers to process data quickly. Less powerful PCs became less important, causing several computing experts to declare "the PC is dead."

In that case, distributive computing could be the savior of both PCs and PC-dependent companies like Intel. Think of the amount of time a typical computer sits idle on a desk. Now imagine if someone could keep a computer busy all the time and harness previously wasted computing power. Legion does just that by allowing a single user on a PC to utilize the resources of all computers on the Legion network -- the benefits of a supercomputer at a fraction of the price. So why hasn't anyone thought of this before now?

There are several commercial file- sharing programs, including the infamous music-swapping Napster, but most are intended to perform a single function. For example, SETI@home is just a screen saver, and Napster is just an interactive database. Legion is different because it was designed from the beginning to integrate all these services into a single interface. Because a computer network is made up of such diverse components, this turns out to be a very complicated problem.

First, all computers on a network are not the same platform, or type of computer. A program that works on an IBM will usually not work on a Macintosh, but Legion has to run on both. Legion also has to be transparent to the user -- the user should not have to supply any information to the program and ideally should not even realize Legion is running. In a distributive process "you don't want to see the feet going on underneath the water. Hopefully you just see the swan gliding along the top," said Sarah Wells, technical writer for Legion's research group.

Other difficult issues include determining who should have access to what information on a network computer, as well as technical computer language problems such as naming variables and sharing computer resources.

Legion is novel because it is a "complete and integrated file distribution system" where all the components "fit together in a clean way," said Prof. Grimshaw. In addition, distributive systems have "no big brother who oversees the whole thing," said Wells. Lisa Romano, Director of Marketing for the commercial Legion software, noted that "once you have the software and permission, you can use another computer without going through anyone else."

Legion was featured in the keynote address at Intel's Peer to Peer conference concerning distributive computing and data management.

The University currently houses a cutting-edge installation of networked PCs running Legion software named Centurion after the class of Roman foot soldiers. It is primarily used to test new versions of the program. In addition, several University professors and graduate students in various departments use Centurion's vast computing capability to perform data processing tasks for their research. Biochemistry and Molecular Genetics Prof. William Pearson used Centurion to compare protein and DNA sequences and was pleased because it expedited his research. "It lets us do several things in a few hours that might take days."

As with most complicated problems, some questions remain. A whole branch of Legion, fault tolerance management, looks for failed processes and redirects tasks to a different computer until the problem is fixed.

"On really large systems, I believe the biggest problem is probably fault tolerance," Grimshaw said. "If you have a Web browser running and a window takes a really long time to load, how do you know when to press the reset button?" Compared to a person, a computer program takes a longer time to stop a failed process or "press the reset button."

"Occasionally system administrators take machines out and although Legion can recover from that, it takes time," Professor Humphreys said. "Sometimes really unexpected things happen and users experience some grief," although Legion eventually will fix the problem, he said.

Seven years ago "when we sat down to design Legion, we tried to think of what the world would be like in 10 years" and plan for it, Professor Grimshaw said. In light of the recent fascination with distributed processing, the computer world seems to be three years ahead of schedule

Comments

Latest Podcast

From her love of Taylor Swift to a late-night Yik Yak post, Olivia Beam describes how Swifties at U.Va. was born. In this week's episode, Olivia details the thin line Swifties at U.Va. successfully walk to share their love of Taylor Swift while also fostering an inclusive and welcoming community.