ABSTRACT
This chapter discusses the fabric that interconnects multiple processors. Two of the most important performance metrics for an interconnection network are latency and bandwidth. A shared memory multiprocessor has unique requirements for interconnection fabric performance, compared to other systems, such as local area networks or the internet, due to the several unique characteristics of communication in a shared memory multiprocessor. Interconnection network glues multiple processors together to form a shared memory multiprocessor system. A processor, or a group of processors, is encapsulated into a node. A node may consist of one or more processor cores, caches or cache hierarchy, memories (M), communication controller, which is the logic that interfaces with the network fabric through a router, which connects to other routers to form the network.
