Algorithms and Analysis of Energy-Eﬃcient Scheduling of Parallel Tasks
For six decades, the concept of the performance of a computer has been equivalent to the computing speed measured by ﬂoating-point operations per second (FLOPS). The peak speed of high-performance supercomputers has increased at an exponential speed.At the same time, thepeakpower requirements also increase at the same rate . To achieve higher computing performance per processor, microprocessor manufacturers have doubled the power density at an exponential speed over decades, which will soon reach that of a nuclear reactor . The emphasis on speed has led to the emergence of supercomputers that consume tremendous amounts of electrical power and produce so much heat that excessive cooling facilities must be constructed to ensure proper operation. Furthermore, the adoption of speed as the ultimate performance metric has caused other metrics such as reliability, availability, and usability to be largely ignored. Consequently, there has been an extraordinary increase in the total cost of ownership of a supercomputer.