ASTRA: a multilayer parallel processing system

The application of parallel processing systems to the solution of computationally intensive problems is a well-established strategy in many application areas. However, in a number of these fields, such as, natural language processing and computer vision, the operations and data structures vary widely in size and complexity so that no single processor is ideally suited to the entire range of problems. Heterogeneous computing environments, composed of diverse high-performance computers connected by high-speed networks have been proposed for these applications. An alternative approach would be to incorporate the principle of heterogeneity in a machine whose topology is compatible with the range of problems. ASTRA, a multilayer parallel processing system developed for image analysis applications, is an example of such a system and its architecture and performance are discussed