OptDef - Extensions for High Performance Computing and Monte Carlo Management

To effectively characterize the Ballistic Missile Defense System (BMDS) using modeling and simulation (M&S), techniques from the fields of simulation optimization, analysis, and uncertainty quantification can be used, but these techniques require the execution hundreds or thousands of simulation runs. OptTek, a pioneer in commercial simulation optimization, initially designed search algorithms for single core and small multi-core environments. These algorithms are very efficient for a small number of parallel simulations, but to take advantage of HPC environments, the algorithms should proceed in a fundamentally different way. OptTek proposes to research, design, code, test, and document new methods that will fully exploit HPC environments to support MDA's M&S mission. We will take simulation optimization methods that are already state-of-the-art for small numbers of parallel simulations, and make them efficient and performant when used with highly parallel evaluations. These new methods are anticipated to yield significant gains in search efficiency. In this effort, we will also add support for dynamic Monte Carlo (MC) run management, providing guidance for analysts in the number of MCs to execute based on the type of analysis being performed, the desired level of confidence, and considering sources of uncertainty besides the modeled stochastic uncertainty. Approved for Public Release | 17-MDA-9395 (24 Oct 17)