Hiding Memory Latency using Dynamic Scheduling in Shared-Memory Multiprocessors