|Volume 7, Issue 1 -
Associate Professor, computer Science and Director, High Performance Systems, Software Laboratory, University of Maryland
Joel Saltz leads a research group at the CRPC's affiliated site at the University of Maryland, College Park whose goal is to develop methods that will make it possible to produce portable compilers that generate efficient multiprocessor code for irregular scientific problems (i.e., problems that are unstructured, sparse, adaptive, or block structured). He collaborates with a wide variety of applications researchers from areas such as computational fluid dynamics, computational chemistry, computational biology, environmental sciences, structural mechanics, and electrical power grid calculation research.
A key aspect of the research on irregular scientific problems is the development of portable runtime support libraries that coordinate interprocessor data movement, manage off-processor data, support a shared name space, and couple runtime data and workload partitioners to compilers. This runtime support is then used in distributed memory compilers by CRPC researchers at Rice University and Syracuse University. The runtime support is also used to port applications codes to a variety of multiprocessor architectures. Saltz is now focusing increased attention on programs that arise from a challenging set of applications that are characterized by data access patterns that are determined at runtime and can also change dynamically and/or inhibit parallelization. These dependency patterns are known only at runtime but can be fully predicted before a program enters an irregular loop or loop nest.
Saltz's group is also extending this work and is developing C++ based class libraries (CHAOS++) for irregular scientific problems. This work is being carried out in collaboration with Los Alamos National Laboratory and the California Institute of Technology. CHAOS++ is being used to develop parallelized application codes and is also being used as part of the runtime support for the University of Indiana's pC++ compiler.
Irregular scientific problems frequently make use of very large data sets. In many cases, each processor will need to carry out irregular accesses to secondary storage. Saltz's group is developing irregular problem runtime support that is able to generate optimized patterns of disk access. Currently, the major driving applications for this work are Maryland's Land Cover Dynamics Grand Challenge project and applications associated with submarine structural acoustics. The development of compiler support for scalable I/O is being carried out in collaboration with the Scalable I/O Consortium and the CRPC.
Saltz came to the University of Maryland after spending three years at ICASE at the NASA-Langley Research Center as lead computer scientist and three years at Yale University as an assistant professor. In 1985, he received a Ph.D. in computer science and an M.D. from Duke University. Among his professional activities, he has edited and contributed to several respected journals.
Table of Contents