Parallel finite element analysis platform for the earth simulator: GeoFEM

Authors:
Hiroshi Okuda;Kengo Nakajima;Mikio Iizuka;Li Chen;Hisashi Nakamura
Affiliations:
Department of Quantum Engineering and Systems Science, University of Tokyo, Tokyo, Japan;Research Organization for Information Science & Technology, Tokyo, Japan;Research Organization for Information Science & Technology, Tokyo, Japan;Research Organization for Information Science & Technology, Tokyo, Japan;Research Organization for Information Science & Technology, Tokyo, Japan
Venue:
ICCS'03 Proceedings of the 2003 international conference on Computational science: PartIII
Year:
2003

Citing 3
Cited 2

Three-dimensional elasto-static analysis of 100 million degrees of freedom

Advances in Engineering Software
Parallel performance optimization of large-scale unstructured data visualization for the earth simulator

EGPGV '02 Proceedings of the Fourth Eurographics Workshop on Parallel Graphics and Visualization
Parallel Iterative Solvers for Unstructured Grids Using an OpenMP/MPI Hybrid Programming Model for the GeoFEM Platform on SMP Cluster Architectures

ISHPC '02 Proceedings of the 4th International Symposium on High Performance Computing

Concurrent Visualization in a Production Supercomputing Environment

IEEE Transactions on Visualization and Computer Graphics
Model-driven autotuning of sparse matrix-vector multiply on GPUs

Proceedings of the 15th ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming

Quantified Score

Hi-index	0.00

Visualization

Abstract

GeoFEM has been developed as a finite element solid earth simulator using the Earth Simulator (ES) (35.61 Tflops/peak according to the Linpack benchmark test). It is composed of a platform and some pluggable 'analysis modules' for structural, electromagnetic thermal fluid, and wave propagation simulations. The platform includes three parts: parallel I/O interface, iterative equation solvers and visualizers. Parallel solvers have got very high performance on the ES. When using up to 176 nodes of the ES, the computational speed of the static linear analysis by the optimized ICCG (Conjugate Gradient method with Incomplete Cholesky Preconditioning) solver, has reached 3.8 TFLOPS (33.7% of peak performance of 176 nodes). Parallel visualizer can provide many visualization methods for analysis modules covering scalar, vector and tensor datasets, and have been optimized in parallel performance. The analysis modules have also been vectorized and parallelized suitably for the ES and coupled on memory with the parallel visualizer.