Related Experiment Videos
A new hierarchical parallelization scheme: generalized distributed data interface (GDDI), and an application to the
Dmitri G Fedorov1, Ryan M Olson, Kazuo Kitaura
1National Institute of Advanced Industrial Science and Technology, 1-1-1 Umezono, Tsukuba, 305-6568 Ibaraki, Japan. d.g.fedorov@aist.go.jp
Journal of Computational Chemistry
|March 11, 2004
Summary
A new hierarchical parallelization scheme, generalized distributed data interface (GDDI), enhances computational chemistry software like GAMESS. GDDI significantly improves performance for tasks such as numerical derivatives and fragment molecular orbital calculations, especially on large parallel systems.
Area of Science:
- Computational Chemistry
- High-Performance Computing
- Parallel Algorithms
Background:
- Standard parallelization schemes can limit computational efficiency for complex molecular simulations.
- Efficiently distributing tasks in large-scale quantum mechanical calculations remains a challenge.
Purpose of the Study:
- To introduce and evaluate the generalized distributed data interface (GDDI), a novel two-level hierarchical parallelization scheme.
- To assess the performance and scalability of GDDI for computational chemistry tasks within the GAMESS software.
Main Methods:
- Implementation of a two-level hierarchical parallelization strategy in GAMESS.
- Application of GDDI to numerical derivatives (geometry optimizations, frequency analyses) and Fragment Molecular Orbital (FMO) method.
- Testing and analysis of GDDI performance and scalability on various network configurations and node counts.
Main Results:
- GDDI demonstrates superior scalability compared to the standard distributed data interface (DDI) scheme, achieving 93 times speedup on 128 nodes.
- High scalability (80-90%) was observed for FMO calculations on 128 nodes for molecular systems like water clusters and proteins.
- Numerical gradient calculations achieved 70% scalability on 128 nodes, showcasing GDDI's effectiveness.
Conclusions:
- GDDI is a flexible and high-performance parallelization tool that effectively utilizes network topology.
- The scheme offers significant speedups and scalability for demanding computational chemistry tasks, particularly on massively parallel computers.
- GDDI is poised to become a preferred method for appropriate computational tasks on large-scale parallel architectures.