Compframe 2005

Atlanta, GA, June 22-23, 2005

Distributed Components for Integrating Large-scale HPC Applications.
Nanbor Wang and Johan Carlsson
Abstract
Component-based software engineering (CBSE) [1, 2] is now a widely accepted software paradigm for developing large, complex enterprise applications. CBSE promotes software reuse and simplifies building and maintaining these large applications by separating different aspects of software development, e.g. component interaction models, component implementations, and component composition, into individual entities that can easily be maintained [3]. There exist many popular enterprise component standards and implementations for different application domains. Examples include Java Beans [4] and ActiveX [5] for building single executables , and CORBA Component Model (CCM) [6], Enterprise Javabeans (EJB) [7] and Distributed Component Object Model (DCOM) [8] for building large-scale distributed applications.
Existing component standards and frameworks, however, are designed with enterprise applications in mind and do not address the needs for High-Performance Computing scientific applications such as combustion modeling, global climate modeling, and fusion plasma simulation. As a result, these existing frameworks do not support features that are important for HPC scientific applications, such as the interoperability with scientific programming languages (FORTRAN) and parallel computing infrastructure (message passing standards, such as MPI). The DOE-sponsored Common Component Architecture (CCA) [9] was developed as a grass-roots effort to address these needs of the HPC communities before they may be adequately served by the commodity off-the-shelf software frameworks and tools [10].
In this paper, we demonstrate the needs for distributed components in building large-scale, distributed and parallel HPC applications. We also present our ongoing work in developing distributed CCA components. Our work is built on top of Babel/Ccaffeine parallel CCA tools and existing remoting technologies. We then review related work. Finally, we present our concluding remarks and sketch the future direction for our research.
Access full paper (pdf file)