ABSTRACT
In single instruction stream, multiple data (SIMD) and multiple instruction stream, multiple data (MIMD) architectures, several processors are connected with each other and memory blocks through the interconnection network, in a tightly coupled manner. These systems typically reside in one or more cabinets in a room rather than widely dispersed geographically. They cannot be easily expanded in small increments and typically employ only one type of processor and hence are not suitable for environments with an array of specialized applications. They usually employ a xed interconnection topology, thereby restricting the users when the applications dictate a different more efcient topology. Some modern SIMD and MIMD systems have addressed these shortfalls by using heterogeneous processing nodes and being scalable to a fairly large number of nodes. They have also merged the SIMD and MIMD concepts, as evidenced by the evolution of the Thinking Machines’ Connection Machine series. The earlier machines in the series were SIMDs while the CM-5 operates in both the modes.
