In-memory integration of existing software components for parallel adaptive unstructured mesh workflows

摘要

Reliable mesh-based simulations are needed to solve complex engineering problems. Mesh adaptivityrncan increase reliability by reducing discretization errors but requires multiple softwarerncomponents to exchange information. Often, components exchange information by reading andrnwriting a common file format. This file-based approach becomes a problem on massively parallelrncomputers where filesystem bandwidth is a critical performance bottleneck.Our approach usingrndata streams and component interfaces avoids the filesystem bottleneck. In this paper,wepresentrnthese techniques and their use for coupling mesh adaptivity to the PHASTA computational fluidrndynamics solver, the Albany multi-physics framework, and the Omega3P linear accelerator frequencyrnanalysis applications. Performance results are reported on up to 16,384 cores of an IntelrnKnights Landing-based system.