An Effective Feedback-Driven Approach for Energy Saving in Battery Powered Systems Duy Le Department of Computer Science The College of William and Mary Williamsburg, VA 23185, USA duy@cs.wm.edu Haining Wang Department of Computer Science The College of William and Mary Williamsburg, VA 23185, USA hnw@cs.wm.edu Abstract—Energy efficiency is essential to battery-powered (BP) mobile systems. However, existing energy efficiency tech- niques suffer from imbalance between system performance and power consumption. This paper presents a Feedback QoS based Model, called FQM, to successfully achieve power reduction without performance degradation. By observing system behavior via control variables, FQM applies pre-estimated policies to monitor and schedule I/O activities. We implement a prototype of FQM under Linux kernel and evaluate its effectiveness with different applications in terms of power consumption, QoS, and performance. Our experimental results show that FQM can effectively save energy while maintaining high QoS stability. Index Terms—Energy, QoS, Feedback Control I. I NTRODUCTION Energy consumption has been a critical issue to battery- powered (BP) computing systems, ranging from laptops to netbooks. The productivity of a BP system heavily depends on its battery runtime. According to [1], with different loads, the runtime of a non-sleep BP system can vary from 0.5 to 5.5 hours. It is a challenging task for these BP systems to achieve high energy efficiency while meeting application quality of service (QoS) requirements [2], [3]. Besides Windows, Linux has become popular for BP sys- tems. As of 2009, Linux is installed in 32% (11 million) of produced netbooks [4]. Under the Linux kernel, new features have been integrated with Intel architectures to save energy [5], including tickless idle and power management (PM) for sub- system components, such as applications, processors, devices, and buses. In particular, Linux kernel version 2.6.23 [6] supports a QoS power management (QoSPM) that enables ag- gressive power management subsystems without substantially affecting the user QoS expectations. In other words, subject to usability and performance constraints, QoSPM creates the possibility of saving energy from the application level. In this paper, we exploit this possibility to address a twofold problem: energy saving with QoS provisioning on BP systems. Based on a control theory analysis, we propose a feedback- driven model, called FQM, to optimize interactions between application and system I/O. Specifically, FQM includes five main components: Source, Executor, Monitor, Controller, and Manager. Its input consists of (1) requests that specify ap- plication utilization, (2) changes of the feedback utilization that is defined as an FQM metric to specify the system CPU utilization, and (3) references of assigned energy and QoS policies for requests to minimize power consumption. The output includes two controlled variables to be monitored: the miss ratio of those requests that do not meet its assigned QoS, and actual CPU utilization. The majority of the FQM prototype is implemented under the Linux kernel. Its internal components, located at the kernel level, handle user I/O transactions, assign QoS parameters for produced requests, and monitor the feedback of control vari- ables. The external components, which are initially built from system behavior observations at the user level, are maintained as utilization-based policies. To validate the efficacy of FQM, we conduct a series of real experiments on a laptop running different applications. Our focus is on three aspects of the laptop: power consumption, QoS, and system performance. The power consumption evaluation indicates that FQM can save a significant amount of energy when multiple applications are active simultaneously. The QoS investigation quantifies the variation of QoS parameters and demonstrates the QoS stability in FQM. Finally, the system performance evaluation shows that the overhead induced by FQM is minor, resulting in energy efficiency for the whole system. In summary, FQM can reduce energy consumption by as much as 20% and maintains QoS stability with high utilization and low miss ratio on the running system. Note that FQM is independent of QoSPM architecture and functionalities, while its prototype works well on QoSPM- based Linux systems. Our FQM design dynamically adapts and guarantees application QoS based on user/system interaction behaviors. This is different from the power-aware systems of Lu et al. [7], [8] which do not aim to guarantee the application QoS, the FCS of Lu et al. [9], [10], [11], which is applicable only for adaptive real time systems, the EC of Minerick et al. [12], which exploits the voltage and frequency of processor to meet a given energy level, the HAPPI of Pettis et al. [13], which automatically simplifies power policies but does not considers all primitive I/Os, the Grace-OS of Yuan et al. [2], which requires a predictable CPU scheduling and only exploits particular multimedia tasks, and QoSPM, which only statically maintains application QoS. The remainder of this paper is organized as follows. In Section II, we categorize and specify the IO-based interaction