Structured Parallelism and Data Access Pattern

In this paper, we start analyzing the vision analysis computational requirements, then we introduce the notion of structured parallelism, and describe the problem of memory data access pattern. Both notions are addressed in the context of parallel distributed memory architecture. Next, the Paralation model is described. Whose essential feature is the clear distinction between computation and communication The design criteria and the current implementation of the visual parallel programming environment are reviewed. The vappe is built around the Paralation computational model. The vappe primary goal is to build an experimental distributed memory parallel programming environment to obtain information on the type and the spatial distribution of the memory data structure accesses for symbolic computations. Finally some considerations on extending the Hierarchical Task Graph program representation to account also the data structures access costs are presented.