Supercomputers-revolution or evolution?

The advantages of a holistic approach to massive parallelism that fully incorporates software requirements are discussed. This idea is considered in the context of a revolutionary approach to achieving practical thousandfold parallelism by the year 2000 that calls for immediately building machines with thousands of processors, using whatever architectural ideas are currently deemed feasible from the system packaging viewpoint. This is contrasted with evolutionary approach, which is to methodically double the number of processors each year, while nurturing and enlarging the software base correspondingly.<
>