Modeling of battery-powered computer systems to enable dynamic power management

With the rapid progress in semiconductor technology, chip density and operation frequency have increased, making power consumption in battery-operated portable devices a major concern. Dynamic power management (DPM), which refers to a selective shutoff or slowdown of system components that are idle or underutilized, has proven to be a particularly effective technique for reducing power dissipation in such systems. In this dissertation, four novel DPM techniques are proposed as summarized next.; i. A battery-aware DPM technique based on continuous-time Markovian decision process model is proposed. This technique addresses the problem of maximizing the battery energy capacity utilization in a portable electronic system under performance constraints. First, a detailed stochastic model of a battery-powered electronic system is presented. Next, the battery-aware DPM problem is formulated as a policy optimization problem and solved exactly by using a linear programming approach.; ii. A hierarchical DPM architecture is proposed that aims to facilitate power-awareness in an Energy-Managed Computer system with multiple components. The proposed architecture divides DPM function into two layers: system-level and component-level. The system-level DPM is formulated as a service request flow regulation problem. In addition, application scheduling is integrated with system-level DPM to achieve higher power savings.; iii. A perturbation analysis based DPM optimization technique is presented to determine the optimal timeout values for an electronic system with multiple power saving states. A Markovian process based stochastic model is described to capture the power management behavior of an electronic system under the control of a timeout policy. Perturbation analysis is used to construct offline and online gradient-based approaches to determine the set of optimal timeout values.; iv. A technique is proposed to combine DPM with dynamic voltage scaling to address the problem of power optimization of a real-time system having heterogeneous components and performing periodic hard real-time tasks. More precisely, an integer programming based formulation is first provided to exactly state the optimization problem to be addressed. Next, a progressive refinement algorithm is presented to solve the power-aware task scheduling and voltage-to-task assignment problems to minimize the total system energy consumption.