Since most shared memory multiprocessor machines only support a small number of processors, where there is minimal benefit from the local memory of the NUMA architecture, UMA is most popular for those machines. However, as the number of processor incr...
Since most shared memory multiprocessor machines only support a small number of processors, where there is minimal benefit from the local memory of the NUMA architecture, UMA is most popular for those machines. However, as the number of processor increases, most large-scale shared memory multiprocessor machines utilize a NUMA structure. In this paper, we present a new cluster cache architecture for a NUMA shared memory multiprocessor system with hierarchical bus which can overcome the limitation of scalablilty of single bus shared memory multiprocessor system. The cluster cache architecture enhances the system performance by reducing both the higher level bus contention and memory access latency.