Alumni Project
Building Component-Based Terascale
Applications Using the CCA
Summary
As the Center for Component Technology for Terascale Simulation Software (CCTTSS) is developing and refining the tools and techniques for component-based high-performance scientific computing, and numerous applications groups are moving to adopt the Common Component Architecture (CCA), a number of pioneering efforts are already using it to create standardized interfaces, reusable components, and full-blown research applications. CCA technology is helping to increase the productivity and capability of simulation-based research efforts. These efforts involve the close cooperation of researchers affiliated with the CCTTSS and a variety of other SciDAC projects, and they draw on a base of other tools and components that represent the work of many others.
Combustion Simulations
The SciDAC CFRFS1 project is developing a component-based toolkit for flame simulations. The vision is a facility where researchers can implement physical/chemical models and numerical algorithms with minimal knowledge of the supporting software infrastructure.
Simulations of flame-like reaction-diffusion systems with replaceable models have been made possible through the use of components for time integration, structured adaptively refined meshes (SAMR), and physical models. Figure 1 shows the evolution of ignition fronts in an igniting H2-Air mixture. Within four months, CFRFS researchers incorporated a second generation of components that embody higher accuracy and stabilized numerical techniques by replacing just a few components in the application. The CCA has also been instrumental in facilitating the use of external software, including components from the SciDAC PERC2 and TOPS3 Centers.
Figure 1. Temperature distribution 0.395 ms after inception of heating. The white lines denote domain decomposition on 28 CPUs.
Forthcoming work includes developing new convective physics capabilities as well as new SAMR numerical schemes to achieve complete flame simulations.
Computational Chemistry
Quantum chemistry (QC) is a major tool in the accurate simulation of chemical phenomena, ranging from combustion modeling to the design of catalysts, pharmaceuticals, and novel materials. QC software packages are typically quite large and complex and not designed for easy interoperability. Two efforts are using CCA technology to achieve new levels of interoperability in this domain.
Researchers at ANL, PNNL, and SNL are linking the NWChem4 and MPQC5 QC codes with the TAO6 optimization package as well as with the linear algebra capabilities within Global Arrays7 and PETSc8 to provide chemists with an unprecedented level of flexibility in determining molecular structures. Figure 2 depicts the components involved. Future work will include interoperability at deeper levels within the QC packages and extensions to complex optimization problems such as protein/ligand binding studies.
Figure 2: Component-based molecular geometry computations. Gray components can be swapped in to create new applications with different capabilities.
We are also collaborating with the SciDAC project Advanced Software for the Calculation of Thermochemistry, Kinetics, and Dynamics (PI: Al Wagner) to develop component-based software to study reaction dynamics.
Climate Modeling
Computational climate modeling is critically important for our understanding of global processes and the potential for human impact. The CCTTSS is collaborating with two major climate modeling efforts in a two-way exchange. With the SciDAC CCSM9 and NASA ESMF10 efforts, work is underway to adapt portions of the Model Coupling Toolkit11 (MCT) to use the CCA environment. Components derived from the MCT will also be used to create a more general paradigm for component-based coupling of multiple simulations. Components are also being introduced into the Community Atmosphere Model (CAM), a part of the CCSM effort, to decouple the “physics” and “dynamics” portions of the code, in order to simplify enhancements to the model and facilitate the scientific collaboration of climate researchers around the code itself.
Scientific Components
A common theme in the above applications is combining application-specific components with more general-purpose ones that can be reused across a range of applications. The CCTTSS is (internally and through collaboration) developing prototype high-performance components and applications, which serve both for education and as production tools for others to use. These components include various services, tools for mesh management, discretization, linear algebra, integration, optimization, parallel data description and redistribution, visualization, and performance evaluation. An important facet of this work is defining domain-specific abstract interfaces, which helps in realizing our vision of plug-and-play scientific components. We collaborate with the SciDAC TSTT12, APDEC13, and TOPS33 centers in this work, and we engage the community at large to participate in similar dialogues in their particular areas of expertise.
Further Information:
Rob Armstrong (PI)
Sandia National Laboratories
(925) 294-2470
Email: rob@sandia.gov
http://www.cca-forum.org
1 A Computational Facility for Reacting Flow Science, PI: H. Najm, http://cfrfs.ca.sandia.gov
2 Performance Evaluation Research Center, PI: David H. Bailey,
http://perc.nersc.gov
3 Terascale Optimal PDE Simulations Center, PI: David Keyes, http://tops-scidac.org
4 http://www.emsl.pnl.gov/docs/nwchem/
5 http://aros.ca.sandia.gov/~cljanss/mpqc
6 http://www.mcs.anl.gov/tao
7 http://www.emsl.pnl.gov:2080/docs/global/ga.html
8 http://www.mcs.anl.gov/petsc
9 Community Climate System Model, PIs: John Drake and Robert Malone, http://www.cgd.ucar.edu/csm
10 Earth System Modeling Framework, PIs: Tim Killeen, John Marshall, and Arlindo da Silva, http://www.esmf.ucar.edu
11 http://www.mcs.anl.gov/mct
12 Terascale Simulation Tools and Technologies Center, PIs: Jim Glimm, David Brown, and Lori Freitag Diachin, http://www.tstt-scidac.org
13 Algorithmic and Software Frameworks for Applied Partial Differential Equations, PI: Phil Collella, http://davis.lbl.gov/APDEC
back to project page
