Untersuchungen von Zugriffen auf den Arbeitsspeicher in Vektorrechnersystemen

Presently, the highest performance computer systems are the vector processors which are mainly employed for technical and scientific applications. However, the performance potential of the pipelined functional units of a vector processor can only efficiently be used if the required data transfer between main memory and the CPU, respectively the CPUs in case of multiprocessor systems, can optimally be routed. This research investigates the access to main memory of a vector processorsystem (single processor System as well as multiprocessor systems with shared memory) with the aid of analytic models, measurements on a double processor System CRAY X-MP and simulations. The emphasis of the investigations is put on lang sequences of accesses to memory cells whose addresses differ from each other by a constant distance as they typically appear in vector processors