SBIR-STTR Award

Management Communications and Control, Inc. (MCCI) proposes specification of applications for high performance scalable, distributed computing systems as "virtual circuits", data flow graph based software architectures. The mathematical properties of data flow, which are similar to physical current flow, enable application of circuit theory based synthesis and design methods to create highly efficient applications for parallel architectures. Proposed methodology innovations will exploit data flow graph properties of equivalency and hierarchy. Equivalent parallel graph segments will be generated for specified architecture partitions creating parallel architectures that readily translate to highly efficient designs. Hierarchical decomposition of architectures design's matching communications loading with hierarchical network capacity will produce application designs with balanced processing and communications loads that is required for high efficiency. The company will demonstrate the capability of its computer aided design and automatic code generation technology, the Autocoding ToolsetT, to support the innovations explored. Extensions and new capabilities that are required to establish full CAD support for developing highly efficient, scalable processing applications will be identified. The methodology innovations we will develop will be incorporated in the company's commercial product, the Autocoding ToolsetT. Extensions we identify will be implemented in our tools. Application developers for large, scalable, distributed parallel systems will enjoy a CAD environment for application development that supports creation of applications that efficiently utilize the next generation of super computing architectures and reduced software costs by an order of magnitude or more.

Keywords:
Software Architecture, Data Flow Graphs, Autocoding, Computer Aided Design, Data Parallelization, High Performance Computing, Software Productiv

Management Communications and Control, Inc. (MCCI) proposes extension of data flow graph programming technology into the non-traditional grid based HPC application domain. Phase I demonstrated the feasibility of developing grid based applications with data flow graph programming, design, and automated code generation tools. The naturally systolic execution behavior of data flow graph grids ideally suits them for specification of recursive equation systems simulating physical phenomena or solution of more abstract equation systems. Significant productivity gains were demonstrated with design and automated code generation tools including the automated import of MATLABr functional specifications. Tool based design methods were applied to achieve high execution efficiencies on small parallel architectures of 12 processors. MCCI proposes to demonstrate implementation of high efficiency grid based HPC applications on large architectures of 256 plus processors and productivity gains based on the use of the company''s data flow tools. The company will restructure its data flow programming environment into public domain and commercial product sets to best support HPC application development in DARPA and other government programs and facilitate establishing their commercial market position.

Keywords:
DATA FLOW, AUTOCODING TOOLSETR, MATLABR,COMPUTING GRIDS, PRODUCTIVITY, COMPUTING EFFICIENCY HIGH PERFORMANCE COMPUTING, SYSTOLIC EXECUTION