PARALLEL PROCESSING ALGORITHMS FOR REAL-TIME SIMULATION

THE NATURE OF EW/C3CM CALCULATIONS LENDS ITSELF TO A PARALLEL PROCESSING APPROACH FOR REAL-TIME COMPUTATIONS. EFFECTIVE PARALLEL PROCESSING, HOWEVER, INTRODUCES ADDITIONAL COMPLEXITIES OF LOAD LEVELING BETWEEN PROCESSING RESOURCES, PROCESSOR CONTROL AND SYNCHRONIZATION, AND MANAGEMENT OF THE "AREA OF FOCUS" SUCH THAT THE SIMULATION PERFORMED IS DYNAMICALLY SCALABLE IN THE LEVEL OF DETAIL CONSIDERED AND COORDINATED TO PROVIDE REAL-TIME RESPONSE. IN THIS PROPOSAL, WE ADVANCE A METHODOLOGY AND ARCHITECTURE FOR HIGHLY EFFICIENT REAL-TIME PARALLEL PROCESSING WHICH EXPLOITS THE SPACIAL RELATIONSHIPS BETWEEN THE ITEMS BEING SUMULATED. PROCESSING ELEMENTS OF A PARALLEL ENVIRONMENT ARE ASSIGNED 3-DIMENSIONAL GEOGRAPHIC AREAS OF RESPONSIBILITY. "LOAD BALANCING" IS ACCOMPLISHED BY MODIFYING EACH NODES GEOGRAPHIC AREA OF RESPONSIBILITY. UNIQUELY THE MANAGEMENT OF THE NODE ASSIGNED AREAS IS DE-CENTRALIZED, THE LOCAL NODE DETERMINING CONJESTION AND NEGOTIATING AREA TRANSACTIONS WITH ITS NEIGHBORING NODES. THIS ROBUST DE-CENTRALIZED LOAD-BALANCING IN TURN ALLOWS THE SIMULATION FIDELITY TO BE CONTROLLED DYNAMICALLY AT THE OBJECT LEVEL WHILE MAINTAINING REAL-TIME RESPONSE. THE PROPOSED ARCHITECTURE FREE FROM BUS OR SHARED MEMORY LIMITATIONS IS IMPLEMENTABLE TO CURRENT TRANSPUTER TECHNOLOGY TO PROVIDE COST EFFECTIVE MASSIVELY-PARALLEL REAL-TIME SIMULATION. ANTICIPATED BENEFITS/POTENTIAL COMMERCIAL APPLICATIONS - THE PROPOSED METHODOLOGY WILL VASTLY EXTEND THE DEGREE OF PARALLELISM PRACTICALLY ACHIEVABLE BY LARGE-SCALE SIMULATION AND CERTAIN CLASSES OF MATHEMATICAL MODELING INCLUDING FINITE ELEMENT ANALYSIS. THE ALGOITHMS ADVANCED MAINTAIN REAL-TIME RESPONSE THROUGH A COMBINATION OF ROBUST LOAD-BALANCING ACCOMODATING TRANSIENT IN PROCESSING NODE LOADING. OTHER POTENTIAL APPLICATIONS INCLUDE MOST CLASSES