This is a proposal for Coarse Data Flow Parallelism Project. This project is sponsored by the HPC Developer Community.
Short Name Coarse Data Flow Parallelism Sponsoring Communities HPC Developer Initial participants Theresa Do Dave Hagerty Paul Hinker Tim Jacobson Shandong Lao Brad Lewis Jeremy Week Description: We propose a project to explore Coarse Data Flow parallelism. With the trend to CMT processors, shared memory parallelism is growing in importance. Data flow parallelism offers the theoretical maximum parallelism. Using a coarse approach provides the benefit of increased parallelism on commodity hardware. To start this project, a package of software named CORD (COarse Region Data flow) will be made available. This package targets linear algebra and has been used to develop highly scalable software for sun performance library on SMPs for many years. The goal of the project will be to support the development of highly scalable data flow parallel programs. This will be done by enhancements to CORD by increasing functionality and the breadth of supported problem domains, through tools to support the development of data flow programs, and by exploring new technology.
