NSF grants $12 million to U of C prof for grid software

By Craig Segall

The National Science Foundation has awarded University computer scientist Ian Foster a $12 million grant to develop a system that will seamlessly link together computers across the world. The system, called a grid, will allow scientists to harness the resources of thousands of computers at once, in essence creating supercomputers distributed across continents.

Foster, along with Carl Kesselman of the University of Southern California, will lead the project called National Middleware Initiative (NMI). NMI will draw together computer scientists to develop software that will erase the gaps between computers it is installed on.

This software, called Middleware, links computers together into a continuous network across which resources ranging from storage space to processing power can be instantly transferred. Middleware

Grid continued on page 2

Grid continued from page one

acts as a uniform protocol between different systems and computer languages, removing barriers that would otherwise make communcation difficult.

“It’s Napster for scientists,” said Foster, referring to the popular music-swapping website. On a grid system, scientists will be able to swap data and any other resources stored on their computers. This ability to access many computers at once can be used to solve problems that would otherwise be unmanageable. Problems that have remained unsolved for lack of processor time or memory become much more accessible.

The SETI@home program, for instance, a screensaver that hunts through radio telescope data for signs of intelligent life, has successfully used this principle to gain thousands of computer years of processing time from computers that would otherwise have been idle. It has become, in essence, the world’s fastest computer.

A grid system, which will allow users to consciously devote their computers to such problems, would be even more powerful. “Communities of scientists and engineers can share resources as they pursue some common goal,” Foster said.

Already, many groups of scientists have been working to build grids linking their computers. Foster has been involved in projects for earthquake researchers who need the power of a grid to run their variable-intensive simulations. Physicists are building grids to help deal with the enormous amount of data gathered by particle accelerators, in which each collision may produce millions of particles. Climatologists envision a grid linking weather sensors and computers to help track the progress of global warming. Indeed, almost any project that involves large numbers of either scientists or data points can benefit from a grid system.

“People increasingly use large collaborative efforts,” Foster said. “ is becoming inevitable.”

Foster, author of a 1997 book which coined the term “grid technologies,” has been involved in the effort to develop working grids since interest in them was first piqued in 1995. At that time, an experiment called I-Way linked 17 networks across the country into a very powerful demonstration grid for about a week.

Since then, efforts to solve the problems involved in creating stable grids have occupied groups around the world. The NMI will coordinate those groups and work to develop common standards and software for grids. “We will take all this technology, integrate it with a bunch of others, and really get it out to the community,” Foster said.

Among those technologies is that developed by Foster’s own Globus project, which is devoted to researching and developing the middleware needed to make grids run. The project’s Globus toolkit seems likely to become the standard for grid programs.

Developing more and better middleware programs is the focus of the NMI. Foster and Kesselman will lead three years of research and development that will integrate the findings of the many groups working on grid research. They plan to release their middleware for free distribution. “We haven’t asserted any control over the intellectual property,” Foster said.

There are challenges ahead. A grid must strike a delicate balance between security and freedom of computer sharing, allowing users to safeguard their data even as they allow access to it. Grids must also be resilient enough to withstand a few faulty computers within them. They must be self-aware enough to direct users to the resources they need, which may be on computers all over the world. And all this complex software has to be easy to use. “The software we develop addresses technical issues in a very large, dynamic, heterogeneous environment,” Foster said.

The grid technology being developed by NMI will most likely be used largely by scientists at first, but Foster suspects business uses will follow. Companies could use grids to increase their processing power or store large amounts of data. IBM has already expressed a strong interest in the technology. “None of this is happening in a vacuum. There’s tremendous industrial activity,” Foster said. “We see a convergence of grid and commercial technologies.”