The Critical Need for High Performance Computing Network Simulation Capabilities: The PETSc DMNetwork Approach

Getnet Betrie, Argonne National Laboratory

Critical infrastructures involve multiphysics, are cross-disciplinary, have multiple interdependencies, and each system is a part of interconnected mega-system. Simulation of an individual infrastructure without accounting for such interdependencies is insufficient to support decision-making. Existing tools are deficient in that they are not generic, either lack coupling or are rigidly coupled, can simulate only steady-state solutions, and lack scalability and uncertainty quantification. Thus, they are not extensible, suitable to understand the impact of one system failure on the other systems, simulate transient nature of the systems, be applied for real-time problems at large spatial and temporal scales, or are not suitable to make informed/risk-based decision. To address these limitations, we are developing a multiphysics, multiscale, and scalable modeling package using PETSc DMNetwork. The package allows simulating networked multiphysics systems that are represented by differential algebraic equations or linear and nonlinear equations, on extreme-scale computers. This talk presents the needs for advanced computing capabilities through a case study of a power-water system interdependency. In addition, it presents the key features of DMNetwork, currently available models in the package and their applications, and future work for modeling other networked systems and handling decision-making capabilities.