In shared memory multiprocessors with a general interconnection network, the program excution time significantly depends on shared memory access latency. Prefetching schemes reduce shared memory access latency by exploiting the overlap of processor computations with data access. Many hardware prefetch shcems have been proposed but those prefetching schemes did not considernetwork contention .
In this thesis, we proposed a new adaptive sequential prefetching scheme which consider network contention. The number of late prefetch increases in case of large network latency. The propsed scheme reduces the number of prefetch in such case.
A program driven simulation is employed to evaluate performance of the proposed adptive sequential prefetching scheme. The simulation result shows that the propsed schems works as other schemes in high bandwith interconnection network, and works better in limited bandwith interconnection network.
