Scalability of an Unstructured Grid Continuous Galerkin Based Hurricane Storm Surge Model

Authors:
S. Tanaka;S. Bunya;J. J. Westerink;C. Dawson;R. A. Luettich, Jr.
Affiliations:
Environmental Fluid Dynamics Laboratories, Department of Civil Engineering and Geological Sciences, University of Notre Dame, Notre Dame, USA;Science and Safety Policy Research Division, Mitsubishi Research Institute, Inc., Tokyo, Japan;Environmental Fluid Dynamics Laboratories, Department of Civil Engineering and Geological Sciences, University of Notre Dame, Notre Dame, USA;Institute for Computational Engineering and Sciences, The University of Texas at Austin, Austin, USA;Institute of Marine Sciences, University of North Carolina at Chapel Hill, Morehead City, USA
Venue:
Journal of Scientific Computing
Year:
2011

Citing 5
Cited 0

Multilevel k-way partitioning scheme for irregular graphs

Journal of Parallel and Distributed Computing
Algorithm 586: ITPACK 2C: A FORTRAN Package for Solving Large Sparse Linear Systems by Adaptive Accelerated Iterative Methods

ACM Transactions on Mathematical Software (TOMS)
Semi discrete discontinuous Galerkin methods and stage-exceeding-order, strong-stability-preserving Runge-Kutta time discretizations

Journal of Computational Physics
Time step restrictions for Runge-Kutta discontinuous Galerkin methods on triangular grids

Journal of Computational Physics
A Performance Comparison of Continuous and Discontinuous Finite Element Shallow Water Models

Journal of Scientific Computing

Quantified Score

Hi-index	0.00

Visualization

Abstract

This paper evaluates the parallel performance and scalability of an unstructured grid Shallow Water Equation (SWE) hurricane storm surge model. We use the ADCIRC model, which is based on the generalized wave continuity equation continuous Galerkin method, within a parallel computational framework based on domain decomposition and the MPI (Message Passing Interface) library. We measure the performance of the model run implicitly and explicitly on various grids. We analyze the performance as well as accuracy with various spatial and temporal discretizations. We improve the output writing performance by introducing sets of dedicated writer cores. Performance is measured on the Texas Advanced Computing Center Ranger machine. A high resolution 9,314,706 finite element node grid with 1 s time steps can complete a day of real time hurricane storm surge simulation in less than 20 min of computer wall clock time, using 16,384 cores with sets of dedicated writer cores.