by Karen Green, NCSA Science Writer
San Diego, CA — The team of physicists known as the MILC (MIMD Lattice Computation) collaboration logs about 1 million CPU hours a year on supercomputers across the U.S. In February 2000, after several months of friendly user access, two members of the team began production simulations on the Alliance’s NT Supercluster at NCSA. Already the researchers – Doug Toussaint and Kostas Orginos of the University of Arizona – have become the largest users of NCSA’s NT Supercluster resources.
“We quickly used up our initial 25,000-hour allocation and had to get that allocation supplemented,” says Toussaint. Performance on the cluster has been about 50 megaflops per node, which is roughly the same as on NCSA’s 190 MHz SGI Origin2000 processors, he notes.
“We are in the happy situation of having codes that are quite portable, so we seem to be continually moving around as time becomes available on different machines,” adds Toussaint. “At this time, we are moving full speed ahead on the cluster.”
The MILC collaboration studies quantum chromodynamics, or QCD, a theory that describes the strongest force in nature – the force that binds together quarks into the protons and neutrons that form the nuclei of atoms. Quarks are held together by gluons, the strongest superglue imaginable. Since quarks and gluons are among the most basic known particles in the universe, understanding the behavior of these building blocks of all matter is a key to answering some of the fundamental questions about the origins of the universe.
To conduct QCD simulations computational researchers use a four-dimensional grid, called a lattice, that spans time as well as the three space dimensions.
The finer the lattice, the more accurate the predictions that result from a QCD simulation. Unfortunately, going to a finer lattice rapidly increases the computer time needed and therefore the cost of the simulation. To get more accurate results from simulations on a relatively coarse lattice, Orginos and Toussaint use procedures known as “improved action,” which are more sophisticated ways of approximating QCD theory on the space-time lattice.
The elementary particles in the strong force interactions are quarks and gluons, which have roles analogous to electrons and photons, respectively.
Simulating the dynamics of gluons is straightforward, but simulating the dynamics of quarks is a very nasty numerical problem. As a result, QCD simulations make compromises in dealing with the quarks, either by omitting, or “quenching,” some or all of the flavors or by calculating at unphysical masses of the quarks. Quarks come in six varieties, or flavors, each with successively greater mass: up, down, strange, charm, bottom, and top. Three of these flavors — up, down, and strange — are light enough to be important in determing the physics of the most common elementary particles.
While many calculations include only two flavors of quarks (up and down), the MILC group’s simulations include a third flavor – the strange quark.
Adding a third quark to the simulation, says Toussaint, more realistically represents the masses and interactions among the quarks.
QCD simulations make for some of the most complex numerical calculations in science, but the team’s code itself is relatively easy to port to a variety of computing platforms. Orginos worked with Rob Pennington and Mike Showerman of NCSA’s NT cluster group to port the code and develop the techniques to run it on the then-experimental NT Supercluster.
“In the end Kostas only had to make one change in the code to avoid one of the variants of the MPI send routines which didn’t work on the cluster,”
Toussaint recalls. “Also, we ported the code very early in the machine’s development. I imagine it would be easier now that the environment has matured.”
The MILC collaboration is a Department of Energy Grand Challenge initiative that involves researchers from nine institutions: the University of Arizona, University of California at Santa Barbara, University of Colorado, Florida State University, Indiana University, University of the Pacific, University of Utah, Washington University, and the Nordic Institute for Theoretical Physics.
Bob Sugar of UCSB is the collaboration’s lead investigator. This research is funded by the Department of Energy and the National Science Foundation.