Most
of the time, the PC on your desk does nothing but wait—for
you to finish reading a Web page, to scroll through your
list of e-mail messages, to think about how you are going
to start a report. And even when your fingers are flying
across the keyboard, you are barely taxing the power of
your CPU. Clearly, a massive amount of computational capacity
goes unused in offices and laboratories around the country.
It is the potential of this computing power that motivates
the proponents of computational grids. Such systems can
meld hundreds of thousands and even millions of computers,
spread across individual local area networks, into a single
virtual computer, capable of handling problems that are
extremely demanding. Grid computing might be used to screen
billions of units of genetic code to identify the genes
and proteins involved in a specific disease. Or a grid approach
might be employed to run a complex engineering simulation,
like those involved in testing wing designs for large aircraft.
In either case, when the grid task is completed, the machines
would be released, available for use by other users who
need the computational power the grid affords but who cannot
justify the costs of their own supercomputers.
Creating a successful framework for computational grids
is no easy task, however. Grids are not simply networks
of computers, but federations of individual users who make
their resources available in exchange for gaining access
to the resources of the grid’s other members. Computer
scientist Martin Humphrey works at the nexus of these organizational
and computational issues, focusing on grid security.
One security issue is authentication. Before users run
applications on the machines in the grid, they may need
assurances that these machines have not been compromised
in any way, putting their application at risk of being stolen.
In effect, each user must make sure that all the computers
in a working grid are what they are supposed to be. At the
same time, the owners of these computers must have some
assurance that the users are indeed who they claim to be.
Humphrey also works on authorization issues, proposing
ways for owners of data to specify access to their resources
among different classes of users without having to qualify
each individual grid member. Or if you are a researcher
who hopes to gain access to a set of resources spread over
many different computers, it would be helpful to have a
system in place so that you can apply to use these resources
without sending out hundreds of requests.
While a number of university researchers have pioneered
grid applications—U.Va. is well known for its Legion
metasystem—corporations like Microsoft and IBM are
also interested in computational grids. “One of the
important challenges of my work right now is to develop
better coordination with commercial developers to ensure
that we are not duplicating our efforts,” says Humphrey.
“We don’t want to solve problems that have already
been solved.” Humphrey is now working on a project
with Microsoft, helping them evaluate if their vision of
the Internet as a vast operating system is consistent with
the work that other grid researchers are doing. Microsoft
certainly appreciates his efforts. This summer, Humphrey
was a featured speaker at Microsoft’s annual Microsoft
Research Faculty Summit.
