Real-time scheduling for pipelined execution of data flow graphs on a realistic multiprocessor architecture

Multimedia and digital signal processing applications have swiftly emerged as important areas for embedded computer systems. Therefore, it is imperative that computer architectures are devised that can cope with the increased burden of processing power implied by these applications. We present a heuristic static real-time scheduling method called real time data flow (RTDF), for mapping data flow graphs onto fixed-size homogeneous multiprocessor architectures. The primary objective of RTDF is to minimize the iteration period of an application. Several other methods with this goal have been proposed in particular in the digital signal processing domain. Similar to many of these, RTDF attempts to minimize the iteration period by exploiting pipelining parallelism. In contrast, however; RTDF assumes an architecture model where communication network characteristics such as message service capacity, communication distance and communication link contention are modeled in a realistic fashion. Therefore, with the aid of RTDF, systems can be designed that support applications with high sample rates without having to resort to dedicated architectures. We show experimentally that RTDF achieves near-optimal iteration periods when communication cost is negligible and that communication overhead only affects the iteration periods of graphs with loops.