A Parallel Processing Paradigm for Irregular Applications

We present a simple and elegant model of computation particularly well-suited to irregularly structured applications in which it is difficult to map the algorithms and data onto a parallel architecture. Unlike other parallel processing paradigms, there is no need to map the application onto the parallel architecture. Nevertheless, an efficient implementation, the Intersecting Broadcast Machine, exists for this model of computation in which the data are mapped entirely at random onto the array of processors. Simulation results for this architecture for a parallel term-matching algorithm indicate high processor and bus utilizations and excellent load balancing characteristics.