Hot spot analysis in large scale shared memory multiprocessors

Scalable multiprocessors that support a shared-memory image to application programmers are typically based on physical memory modules that are distributed. Consequently, the access times for a particular processor to various parts of physical memory differ. The authors explore the implications of this nonuniformity in memory access times. In particular, the study the effect of hot-spots in hierarchical large scale NUMA multiprocessors. They have developed an analytical model of access latencies and contention for shared resources in the interconnection network that links the processors and memory modules. The objective is to provide a better understanding of nonuniform memory access times in scalable architectures. They show the extent to which a variable can be shared before it becomes a performance bottleneck, and assess the potential gain from replication of shared data items. They also demonstrate that the backoff value (after a memory request rejection) must be chosen carefully to balance memory access time and network utilization and that memory utilization is improved by allowing memory request buffering.