Invention DescriptionThe invention provides a method to compute the energy optimal speed profile of an arbitrary device for a generic task. The inputs to the method are the (i) power-speed relationship of the device (obtained by measurement or from the device data sheet), (ii) energy overheads incurred in changing speeds and power modes, and (ii) the performance requirement specified as the amount of work to be done by the device and the deadline by which it must be done. The output of the method is a description of the speed as a function of time (the speed profile) that the device must use to minimize its energy consumption. The only restriction is that the power-speed relationship be a W-convex function (which is a discrete version of a well-known mathematical function called the convex function). The method can handle both discrete and continuous speed sets.Advantages The main advantage of this method is that it does not require an explicit analytical power model of the device, but can compute the exact solution using only the power-speed values corresponding to the device operating points. For example, for a digital integrated circuit with Dynamic Voltage and Frequency Scaling (DVFS) capability, there exist methods to find the optimum speed that use complex analytical power models to account for various types of static and dynamic power consumption. Such models need to be calibrated for each device, and the models themselves need to be changed as technology scales. The proposed method, however, requires no such model or curve fit, and provides a technology independent solution to a variety of speed control problems, and in particular, for DVFS. The final solution is very simple (yet optimal) and is easy to implement as a run-time energy management policy. Due to the generality of the method, it can address a wide range of applications with the same approach, and can thus be incorporated into a system level energy exploration/optimization tool.ApplicationsEnergy optimal speed control of the supply voltage, clock frequency and other dynamically controllable parameters of digital integrated circuits, angular speed of the spindle motor of a disk drive (hard drive or optical drive) during multimedia playback, a network card's radio power and transmission rate for streaming video and sensor networks, and controllable parameters of the cooling mechanism (like fan speed, AC air flow rate, pump pressure, etc.) for thermal management of electronic systems.These are only representative applications, and the general principle can be applied to many other systems.
|Original language||English (US)|
|State||Published - Jun 24 2005|