 |
Center for Nonlinear Studies |
In this talk we will address the design of general purpose scalable pde frameworks for the petascale and eventually exascale solution of fluid-structure interaction problems using task-based approaches. The task-based approach to parallel scientific computing has been proposed as a potential candidate, named the silver model, for exascale software, but is not so often employed at large scales on parallel architectures as of yet. The central idea is to use a Directed Acyclic Graph (DAG) based approach to express the structure of the underlying software. With this in mind, the intention is to explore the usefulness of DAG based approaches, using recent developments in the parallel Uintah software framework for partial differential equations to assess how well the DAG type approach works on present-day large-scale architectures for complex multi-physics multiscale adaptive mesh applications. In particular we will consider both the scalability of the infrastructure and of the underlying algorithms in the adaptive mesh fluid-structure interaction solver. As a result of these investigations, a preliminary and tentative evaluation of the potential of the DAG type approach as a design for PDE software infrastructures at petascale and exascale will be given. The conclusion is that these approaches show great promise for petascale and perhaps eventually exascale but that considerable challenges remain.
Host: Dana Knoll, T-3, 665-8905