PARALLEL SOFTWARE VISUALIZATION AND AUTOMATED PARTITIONING

IN SEARCH OF COMPUTATIONAL ADVANCES, METHODS AND SOFTWARE FOR EASING THE TASK OF DEVELOPING EFFICIENT PROGRAMS FOR SCALABLE PARALLEL COMPUTERS ARE NEEDED. A MUCH HIGHER LEVEL OF COMPUTER ASSISTANCE IS REQUIRED, PARTICULARLY IN PARALLELIZATION AND OPTIMIZATION. EFFECTIVE PARALLEL PROGRAMMING TOOLS NEED TO EASE THE BURDEN OF PARTITIONING A PROBLEM ONTO PARALLEL PROCESSORS, PROVIDE A PORTABLE REPRESENTATION BASIS FOR ALGORITHMS, AND PRODUCE EFFICIENT EXECUTABLE PARALLEL PROGRAMS. CURRENT APPROACHES TO PARALLEL LANGUAGES WHICH PRODUCE NON-PORTABLE PROGRAMS, AUTOMATIC EXTRACTION OF PARALLELISM FROM SEQUENTIAL CODE WHICH PRODUCES LIMITED SPEEDUP DUE TO MISSING DATA ACCESS KNOWLEDGE, AND EXTENDING SEQUENTIAL LANGAUGES WITH DATA PARTITIONING SPECIFICATION WHICH STILL REQUIRES MANUAL PARALLEL PARTITIONING. THE PROPOSED VISUALIZATION AND AUTOMATED PARTITIONING APPROACH TARGETS THE THREE GOALS OF EASING THE PARALLELIZATION BURDEN, PORTABILITY, AND PRODUCTION OF EFFICIENT PARALLEL CODE. IT UTILIZES THE CURRENT STATE-OF-THE-ART SEQUENTIAL LANGUAGES WITH PARALLEL DATA STRUCTURES TO CAPTURE THE PROBLEM DIMENSIONS THAT HAVE THE POTENTIAL TO BE MAPPED ONTO SCALABLE PARALLEL TOPOLOGIES. THIS IS AUGMENTED BY AN INNOVATIVE PARALLEL PROBLEM VISUALIZATION ENVIRONMENT AND AUTOMATED PARTITIONING TOOL TO STREAMLINE THE PARALLEL SOFTWARE DEVELOPMENT PROCESS AND PRODUCE EFFICIENT PROGRAMS FOR DIFFERENT PARALLEL COMPUTER ARCHITECTURES AND SCALABLE TOPOLOGIES. ANTICIPATED BENEFITS: THE SUCCESSFUL COMPLETION OF THIS EFFORT WILL MEET A NEED OF NATIONAL IMPORTANCE FOR THE EFFICIENT USE AND IMPROVED PROGRAMMABILITY OF HIGH PERFORMANCE COMPUTERS. POTENTIAL USERS INCLUDE MANY GOVERNMENT AGENCIES, UNIVERSITIES, AND BUSINESSES WITH SCALABLE AND LARGE-SCALE EXPERIMENT COMPUTING NEEDS IN AREAS SUCH AS BIOMEDICAL ANALYSIS, GEOLOGICAL ANALYSIS, AND PHYSICS MODELING.