l AN HPF BINDING FOR MPI
Argonne National Laboratory and Syracuse University are working 
together to develop mechanisms that allow the coupling of multiple
HPF tasks to form task-parallel computations. We demonstrate a 
library that allows HPF tasks to be coupled by using functions 
defined in the standard Message Passing Interface. Potential 
applications for this library include multiblock codes, 
multidisciplinary simulations, and pipeline algorithms. 
Ian Foster, David Kohr, Rakesh Krishnaiyer, Alok Choudhary

2 WATER SATURATION IN A HETEROGENEOUS OIL RESERVOIR
Researchers at the Center for Subsurface Modeling (CSM) at the 
University of Texas investigate the use of high-performance 
parallel processing to simulate the behavior of petroleum reservoirs 
and groundwater aquifers. Here water has been injected into an oil 
well (foreground) and is spreading down and outward under the 
influence of gravity. In the background, four production wells are 
producing a mixture of oil and water. Phil Keenan

3 DSMC SIMULATION OF DUAL RCS JET INTERACTION
Researchers at the High Performance System Software Laboratory 
at the University of Maryland have collaborated with NASA Langley 
to parallelize the DSMC particle simulation technique on massively 
parallel architectures such as the Intel Paragon and IBM SP-2. 
Applications of the DSMC technique include simulation of dual RCS 
(Reaction Control System) jet injection (shown here) and corner 
flow simulation. Bongki Moon

4 THE D SYSTEM 
Rice University is working on the D System, a suite of prototype 
tools that will support development of programs in Fortran D, an 
abstract, machine-independent parallel programming language. The 
D System prototype tools include an intelligent editor for Fortran D 
and an automatic data distribution assistant. The Fortran D editor 
includes a performance analysis subsystem developed in collaboration 
with the University of Illinois. Vikram Adve, Alan Carte, Gil Hansen, 
Ken Kennedy, Chuck Koelbel, John Mellor-Crummey

6 MESH ADAPTIVITY NEAR AN OIL WELL 
Here the Rice Unstructured Flow Code (RUF), a C++ implementation of 
mixed finite element methods on general geometries for single phase 
flow problems with full tensor coefficients, is used to generate a 
sample 2-D grid that illustrates mesh adaptivity near wells (top) 
and the corresponding pressure field (bottom). Phil Keenan

7 HIGH PERFORMANCE FORTRAN (HPF) 
High Performance Fortran (HPF) is an extension of Fortran 90 that 
gives scientists and engineers a portable, efficient language for 
parallel programming. Its main features are data mapping (laying 
out data across the processor memories) and data parallelism 
(concurrent operations on array elements). HPF was designed by a 
coalition from industry, academia, and government led by CRPC Director 
Ken Kennedy. Ken Kennedy, Charges Koelbel, Mary Zosel, and the many 
members of the High Performance Fortran Forum

8 INPUT/OUTPUT (I/O) OPERATIONS FOR A HARTREE QUANTUM CHEMISTRY CODE
Researchers in the Scalable I/O initiative are working to characterize 
the I/O behavior of large application codes and to disseminate these 
characterizations to the HPCC community. This example depicts the 
beginning of the steady-state, I/O intensive phase of a Hartree quantum 
chemistry code on 128 processors of the Intel Paragon XP/S, with file 
read operations (green), broadcast information to other processors (blue), 
and periodic file seeks (orange). Dan Reed, Will Scullin, Ruth Aydt, 
Stephen Lamm, Andrew Chien, Evgenia Smirni, Phillis E. Crandall

http://www.crpc.rice.edu/CRPC/Images/Panels/captions-1.txt