Techniques in Modeling and Simulation of Multi-Domain Operations
Subtitle: Running many test cases on High Performance Computers (HPCs)

Donnelly, Bob (GDIT)

Category:
Intersection of Digital Engineering and High Performance Computing/High End Computing

Multi-Domain Operations span space, cyber, air, sea, and land. Deploying capabilities across these domains involves variable conditions: terrain, weather, communication links, sensor views of targets, types and locations of targets, and many others. Besides the inherent variabilities of each mission, there are various acquisition and design variables. The acquisition community needs to answer: How would new capabilities impact mission success? How many and in what manner should this new capability be used? How do we optimize its deployment? How do we compare various proposals and designs of new combat vehicles or new satellites? Modeling and simulation provide a method to span these variables, both across those inherent in the domain of operations (weather, etc.) and enabling comparison of alternatives. While running independent test cases are "embarrassing parallel", efficient deployment on HPCs requires analysis and approaches that distribute load across HPC nodes. This talk presents methods of distributing that load. The described variability results in variation in computational needs (memory, CPU, elapsed time) across all runs. For example, one run may take a few minutes and another a few hours. Pre-assignment of runs can result in many HPC nodes idling while a few finish longer calculations. The presented methods allow fully dynamic task assignments to all assigned HPC assets.