Early Communication Expectation Maximization (ECEM). one of the parallel image reconstruction techniques (1,2,3,4), introduction was intended to speeding up the generation process of Positron Emission Tomography (PET) images. Despite the fact that it has been shown to be effective running on distributed memeory parallel machine like IBM SP2 parallel machine, implementation independent issues still remain untouched in designing a better software architecture. In this study, we will characterize the properties of the ECEM application algorithm and try to find the objects that best describe the data structures of our particular scientific application. Abstract data typing techniques will be employed for specifying the framewrok of the application in both complete and precise manner. Message Passing Interface (MPI) primitives also will be embedded into the object model as our target programming paradigm. Parallel I/O activities also are perfored using derived data types with the support of the underlying parallel file system. Being specified not by its object oriented language implementation, the design is used to perform the experiments on IBM SP2 parallel machine with MPI parallel library functions.
展开▼
