A new development framework based on efficient middleware for real-time embedded heterogeneous multicomputers

Achieving application software portability is challenging for a computer-based system that has a single embedded processor. Such portability is even more challenging when performance requirements demand distributing and parallelizing the application software to run on an embedded scalable heterogeneous multiprocessing hardware target. Application software development for such an embedded multicomputing target is just plain hard enough-without even having to consider portability and technology insertion issues. One feasible approach to solve both problems is by using middleware-based technology and complementary design methodologies and tools. One recent approach that is proving viable and effective is the Talaris application configuration middleware framework with an application development tool that layers on top of it known as PeakWare for RACE. Applying these concepts to the embedded multiprocessor signal processing domain is both novel and promising. The middleware and tool are introduced and then discussed being illustrated by a simplified STAP (space-time adaptive processing) application that was rapidly prototyped and run on fifteen compute nodes.