Power Allocation and Task Scheduling on Multiprocessor Computers with Energy and Time Constraints

Performance-driven computer development has lasted for over six decades. Computers have been developed to achieve higher performance. As of June 2010, three supercomputers have achieved petaflops speed: Cray Jaguar (224,162 processors, 1.759 petaflops), Dawning Nebulae (120,640 processors, 1.271 petaflops), and IBM Roadrunner (122,400 processors, 1.042 petaflops) (1). According to Moore’s law of computing hardware, the following quantities increase (decrease) exponentially, doubling (halving) approximately every 2 years: the number of transistors per integrated circuit (cost per transistor), processing speed, memory/storage capacity (cost per unit of information), and network capacity (2). While performance/cost has increased dramatically, power consumption in computer systems has also increased according to Moore’s law. 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 (3). Such increased energy consumption causes severe economic, ecological, and technical problems.