A control architecture for managing instructions among partitions of a data parallel structure

The problem considered is the definition and evaluation of control strategies for handling image features spanning over several partitions of a parallel data structure, whenever such partitions are assigned to different nodes of a distributed memory MIMD parallel machine. The solutions proposed are based on a tree structure of MANAGER processes, which handle communication of information between WORKER processes operating upon the parallel data structure in a distributed way, so avoiding the need of a centralized controller. Even if the structure is not tailored to a particular data parallel algorithm, it is applied (as an example) to a parallel image processing algorithm for edge detection.<
>