Iso-power-efficiency: An approach to scaling application codes with a power budget

For many applications, speedup saturates and parallel efficiency decreases if the problem size is held fixed while increasing the number of processors. For some problems, it is possible to maintain a fixed parallel efficiency by increasing both the problem size and the number of processing elements. The rate at which the problem size must increase to maintain constant efficiency for a given rate of increase of the number of processors is given by the iso-efficiency function. We have developed a new scalability function called iso-power-efficiency that determines the rate at which the problem size must increase to maintain constant efficiency for a given rate of increase in the application\u27s power budget. For a given power budget, an application can choose to use a larger number of processors running at lower power. We show that such overprovisioning can lead to better scaling behavior