Optimization Strategies Using Hybrid MPI+OpenMP Parallelization for Large-Scale Data Visualization on Earth Simulator

Authors:
Li Chen;Issei Fujishiro
Affiliations:
Institute of Computer Graphics and Computer Aided Design, School of Software, Tsinghua University, Beijing, P.R. China 100084;Institute of Fluid Science, Tohoku University, Sendai, Japan 980-8577
Venue:
IWOMP '07 Proceedings of the 3rd international workshop on OpenMP: A Practical Programming Model for the Multi-Core Era
Year:
2007

Citing 8
Cited 1

Display of Surfaces from Volume Data

IEEE Computer Graphics and Applications
Imaging vector fields using line integral convolution

SIGGRAPH '93 Proceedings of the 20th annual conference on Computer graphics and interactive techniques
MPI versus MPI+OpenMP on IBM SP for the NAS benchmarks

Proceedings of the 2000 ACM/IEEE conference on Supercomputing
Flow Visualization with Surface Particles

IEEE Computer Graphics and Applications
Visualizing Second-Order Tensor Fields with Hyperstreamlines

IEEE Computer Graphics and Applications
The stream polygon: a technique for 3D vector field visualization

VIS '91 Proceedings of the 2nd conference on Visualization '91
Evaluation of Cache-based Superscalar and Cacheless Vector Architectures for Scientific Computations

Proceedings of the 2003 ACM/IEEE conference on Supercomputing
Communication Bandwidth of Parallel Programming Models on Hybrid Architectures

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

Enabling low-overhead hybrid MPI/OpenMP parallelism with MPC

IWOMP'10 Proceedings of the 6th international conference on Beyond Loop Level Parallelism in OpenMP: accelerators, Tasking and more

Quantified Score

Hi-index	0.00

Visualization

Abstract

An efficient parallel visualization library has been developed for the Earth Simulator. Due to its SMP cluster architecture, a three-level hybrid parallel programming model, including message passing for inter-SMP node communication, loop directives by OpenMP for intra-SMP node parallelization and vectorization for each processing element (PE) was adopted. In order to get good speedup performance with OpenMP parallelization, many strategies are used in implementing the visualization modules such as thread parallelization by OpenMP considering seed point distributions and flow features for parallel streamline generation, multi-coloring reordering to avoid data race of shared variables, some kinds of coherence removal, and hybrid image-space and object-space parallel for volume rendering. Experimental results on the Earth Simulator demonstrate the feasibility and effectiveness of our methods.