CCS Institutes: MRI Background: Next Steps

CCS Resources

Computers

Contact CCS

CCS Institutes

Communities

Next Steps

When fully operational, the main product of CCS-MRI will be an open source repository of computational tools that is highly optimized on the available HPC resources and that is sufficiently flexible to be readily extended to new areas of research and discovery. Building such a repository of CMS tools from scratch is clearly an enormous project. Furthermore, because many of the tools will have to earn acceptance in the CMS community, it is not clear that such an ab intio approach will pay off in terms of widespread use. Therefore, a different strategy will be adopted.

Initially, a repository will be developed in areas where scientists affiliated with the CCS-MRI have an established track record. In particular developments will be stongly coupled to the already established (and DOE-BES funded) Computational Material Science Network (CMSN) as well as the Center for Nanophase Materials Science that is being developed at ORNL. At the same time the development of a CMS tool set will be initiated using modern object oriented and generic programming techniques through encapsulation of existing elements of major codes. Research projects in key areas of CMS will lead the way for the development of the new algorithms and tools. The staged approach that is outlined below is based upon close integration - rather than the traditional separations - of researcher from the respective disciplines of CMS, computer science, and applied mathematics:

Step 1: Joint scientific research projects with CMS community will provide the close contact needed to assure that the repository and tools developed in steps 2 and 3 meet the goal of being relevant to current materials science and condensed matter research. At the same time this strategy will assure fast introduction of the tools into the relevant user community. Furthermore, CCS-MRI will be linked with two other DOE funded projects, the Computational Materials Science Network (CMSN), and the Center for Nano-phase Materials Science (CNMS). Both of these projects implement models of collaborative research between multiple universities and national laboratories and have a strong need for CMS applied at the nano-scale. To make reciprocal use of these networks and centers, a number of research collaborations with several of the groups involved will be established with joint postdocs and students, which will be supported partly by CCS-MRI and either CMSN or CNMS. In combination CCS-MRI, CMSN, and CNMS will cover by far the largest community of computational materials scientists in the US and rank among the largest and most important internationally. It can also be anticipated that these joint research projects will be the primary source of inspiration and motivations for new algorithmic developments. Research projects will be chosen primarily to address the major problems outlined in the scientific rational section and that are central to the scientific research goals of CMSN and CNMS.

Step 2: A repository of open source CMS codes will be built. A selection of codes that have been developed by groups affiliated with CCS-MRI and have received wide recognition in the CMS community will be adapted to be suitable for wide distribution and entered into an open source repository. Mechanisms for maintenance of scientific codes in the open source domain will be developed, which include. (1) Implementing automated testing procedures with which quality of future releases is assured. (2) Implementing proper revision control and release mechanism to assure that the latest releases are available to the community. (3) Maintenance of up-to date documentation. (4) Organize feedback from the community (bug reports and feedback for future developments).

Step 3: A reusable, extensible, and portable tool-set for CMS will be developed. The primary aim is to reduce the amount of work needed to implement new application codes for CMS while keeping the resulting code at a high level of efficiency. The tools will be designed with modern object oriented and generic programming techniques and organized in a layered manner emphasizing a minimum of interdependence. While some of the tools will be built from scratch, many will be encapsulations of legacy codes. A procedure will be developed to automatically analyze, adapt, and encapsulate legacy codes for the tool-set. In this way, source codes that are still under development elsewhere can be linked into the tool-set with a minimum of maintenance.

Step 4: Tools for code analysis, maintenance and optimization will be developed to make optimal use of the vast legacy of CMS software. Implementation of this step will be based on building a suite of tools that automatically analyze source codes and produce new source code that is either more efficient, better suited for encapsulation into object orient environments, or that reduce the total size of codes in the repository though discovery and removal of redundancy will be developed. These tools will also be useful to maintain legacy software systems to assure portability and efficient use on modern HPC architectures will be developed.

Step 5: Standardization of interfaces between modules and codes as well as the development tools for data analysis and visualization to improve efficiency of the research process. Many of the current limitations to data transfer between methods and length scales can easily be removed with the definition and adherence to standardized interfaces. For existing methods and codes standards will be developed for input and output files. Furthermore, codes in the repository will be adapted to these standards and new algorithms and implementations will only be developed within the context of these standardize interfaces.

Step 6: Integration of tools and interfaces with DOE sponsored computer science projects to enhance performance and applicability of the CCS-MRI developments and to leverage these investments.. As an example the tools developed in step 3 will be transformed into components that can be used in the common component architecture (CCA) and thus automatically make use of modern paradigms for massively parallel computer architectures. Availability of standardized interfaces will simplify integration of the CCS-MRI software with commercial or externally developed data analysis and visualization packages as well as problem solving environments.

Step 7: Maintain HPC resources commensurate with algorithms and methods that grow out of the research projects of CCS-MRI. Developing tools that scale favorably with system size and enable timely development of new models is only one part of the equation. The other is, that computational requirements grow in concert. CCS-MRI will continue to obtain world-class computational facilities to meet the demands of simulating systems with higher resolution, more accuracy, and for longer times. CMS codes make the most efficient use of present day supercomputers and are thus guaranteed to be the first applications to break the 10, 100, and 1000 Teraflop barriers on general purpose HPC machines.

Step 8: Finally, but no less important than any of the previous steps, CCS-MRI will regularly host workshops, seminars, and summer schools. This will include, an already existing, biweekly CMS seminar at ORNL, an annual CCS-MRI sponsored conference with international visibility, as well as a number of targeted workshops for design, implementation, and research teams. The CCS-MRI will also develop web-based training courses and research courses for CMS to propagate the tools developed under CCS-MRI into the research community.

CCS: Institutes: MRI: Background: Next Steps

Next Steps