Submitted by:

Data ow architecture as a concept has been around since the 1970s for parallel com-putation. In data ow machines, the hardware is optimized for ne-grain data-driven parallel computation. The architecture can tolerate long unpredictable communication delays and supports generation and coordination of parallelism directly in the hard-ware. We propose a new coarse-grained data ow model of computation for parallel architectures named Predictive Block Data ow (PBD) due to its predictive capacity and processing of instructions in blocks instead of the ne grain model of one machine instruction. We have simulated the high level machine architecture and implemented two algorithms, quicksort and Dijkstra's single source shortest path, to display the high degree of parallelism that our model can exploit in both highly recursive and irregular data structures.