Scheduling of periodic time critical applications for pipelined execution on heterogeneous systems

Allocation of tasks of a periodic, time critical application, onto a set of processors, with task duplication is examined here. Pipelined execution, where computations and communications take place simultaneously, is employed to meet the required flow rate of the application. The prime objective of the algorithm is to reduce the number of pipeline stages and/or the pipeline period and thereby reducing pipeline latency. The algorithm is well suited for both homogeneous and heterogeneous systems and it is scalable. The order of complexity of the algorithm is O(/spl nu//sup 2/), where /spl nu/ is the number of nodes in directed acyclic task graph (DAG) representing the application. Simulations have indicated that the proposed algorithm generates 9%-52% shorter pipelines.