Dynamic Task Scheduling for Scalable Parallel AMR in the Uintah Framework

Uintah is a computational framework for fluid-structure interaction problems using a combination of adaptive mesh refinement(AMR) and MPM particle methods. Uintah uses domain decomposition and a task graph based approach for asynchronous communication and automatic message combination . The original task scheduler for Uintah ran computational tasks in a predefined order. To improve the performance of Uintah for petascale architecture, a new dynamic task scheduler allow better overlapping of the communications and computations is designed in this study. The new scheduler supports asynchronous, out of order scheduling of computational tasks by putting them in a distributed directed acyclic graph(DAG) and isolating task memory. The effectiveness of this new approach is shown on large scale fluid-structure examples through an analysis of the performance of the software. Dynamic Task Scheduling for Scalable Parallel AMR in the Uintah Framework Qingyu Meng, Justin Luitjens, Martin Berzins {qymeng, luitjens, mb}@cs.utah.edu School of Computing, University of Utah, Salt Lake City, Utah May 10, 2010 Abstract Uintah is a computational framework for fluid-structure interaction problems using a combination of adaptive mesh refinement(AMR) and MPM particle methods. Uintah uses domain decomposition and a task graph based approach for asynchronous communication and automatic message combination . The original task scheduler for Uintah ran computational tasks in a predefined order. To improve the performance of Uintah for petascale architecture, a new dynamic task scheduler allow better overlapping of the communications and computations is designed in this study. The new scheduler supports asynchronous, out of order scheduling of computational tasks by putting them in a distributed directed acyclic graph(DAG) and isolating task memory. The effectiveness of this new approach is shown on large scale fluid-structure examples through an analysis of the performance of the software.Uintah is a computational framework for fluid-structure interaction problems using a combination of adaptive mesh refinement(AMR) and MPM particle methods. Uintah uses domain decomposition and a task graph based approach for asynchronous communication and automatic message combination . The original task scheduler for Uintah ran computational tasks in a predefined order. To improve the performance of Uintah for petascale architecture, a new dynamic task scheduler allow better overlapping of the communications and computations is designed in this study. The new scheduler supports asynchronous, out of order scheduling of computational tasks by putting them in a distributed directed acyclic graph(DAG) and isolating task memory. The effectiveness of this new approach is shown on large scale fluid-structure examples through an analysis of the performance of the software.