|
Features: Nov. 12, 2001
Modeling Random Events
 Researchers
are developing novel computational techniques that could
lead to better simulation of complex systems, such as the
spread of diseases, the evolution of financial markets,
and the flow of Internet traffic.
With a $450,000 grant from the National
Science Foundation, Gyorgy Korniss, assistant professor
of physics, will use a computational technique called Parallel
Discrete-Event Simulation (PDES) to model large-scale systems,
where events occur randomly in space and time.
What makes Korniss' work unique is that
he uses naturally occurring systems to help understand and
develop the sophisticated algorithms necessary for this
modeling process. The physical surface growth of crystals,
for example, can be likened to the evolution of events in
a large class of other systems because they have similar
asynchronous, or random, characteristics. Comparing these
natural systems with the simulated time horizon developed
by the researchers can help explain how advanced algorithms
work and how they can be optimized for the large-scale systems.
What makes Korniss' work unique is
that he uses naturally occurring systems to help understand
and develop the sophisticated algorithms necessary for
this modeling process.
|
Because these systems are so large, the
modeling and simulation would normally be a slow process.
Korniss and his colleagues can speed up the process by breaking
down the system and distributing it over many processors.
Once the system is broken down, the challenge
lies with accurately preserving the random nature of the
evolution of the physical system being modeled, said Korniss.
In order to accomplish this, sophisticated algorithms are
used to program each processor to simulate a random time
stream. When linked together, these streams constitute the
simulated time horizon for the entire system.
Korniss is working with Mark Novotny in
the department of physics and astronomy at Mississippi State
University and collaborating with researchers at Lucent
Technologies.
|
