The purpose of the proposed research is to incorporate predictability into the performance of embedded aircraft data systems implemented by multiprocessor artchitectures. Several characteristics required in aircraft control and associated signal processing include repeated execution of the same algorithm, highly predictable and reliable performance, and the ability to meet hard deadlines for outputs. Dataflow computational models are used to create a real-time multiprocessing design methodology which can achieve performace bounds given sufficient processors. A key approach to achieving predictable performace is to control injection rate of input data and to modify the dataflow graph precedence so that a processor is always available to execute an enabled graph node. In the proposed research effort, investigations will be carried out to achieve predictability under variable task execution times and a scheduling software tool will be developed. Computational times of tasks in aircraft data systems can vary due to cache miss in the processor, data dependent processing, or sharing of common resources such as memory. Predictability of aircraft control algorithms will be absolutely essential for safety and efficiency. Commercial Applications:The schedulting software tool developed will be directly impacting the aircraft industry's cockpit control systems. Commercially available digital signal processing software COMDISCO can license the software from UniSoft and incorporate into their system. Aircraft data system manufacturers like Westinghouse, Loral, Harris and Allied Signal will be able to use and incorporate this tool into their systems. These systems will add value to commercial aircraft companies like Boeing, McDonnel Douglas and Lockheed and aid the pilot in making the right decision.
