Dr. Baris Taskin
Battery lifetime is the primary complaint of smartphone users in surveys and interviews. The increasing capabilities of mobile devices, coupled with the slow growth of battery capacity, demands new energy efficient designs and algorithms. Over the past few decades, researchers have built a large body of research focusing on conserving energy in mobile environments. Most prior work has optimized energy consumption of single components in isolation (e.g. CPU, memory, or network), while a very few efforts have focused on system-wide energy conservation.
Existing power management approaches suffer from multiple problems; approaches that use rate limiting harm performance, are not portable to different devices, and sometimes result in higher energy consumption. Approaches that use Dynamic Voltage and Frequency Scaling (DVFS) to manage energy, track optimal frequency settings and therefore are expensive, new metrics are needed to evaluate the energy efficiency of these systems. This work provides research contributions in four important aspects to improve energy management of mobile devices: introduces two new metrics for energy management, study system-level tradeoffs with multi-component DVFS, an implementation of hardware-software power management for Android. Aspect 1: The research proposes a new metric ”inefficiency” that can be used to specify energy constraints. Inefficiency is both application and device independent unlike existing metrics. Aspect 2: The proposed research explores new and existing hardware capabilities that provide system-level energy performance trade-offs. To this end, trade-offs provided by the DRAM frequency scaling when coupled with CPU DVFS are studied. A simple clustering technique is proposed that can reduce the overhead of managing frequency scaling of these two components. Aspect 3: A full system design is proposed that uses inefficiency as a constraint and can simultaneously manage energy consumption of multiple hardware components dynamically during the execution of the applications. Aspect 4: A set of new metrics, ”power agility metrics” are proposed. These metrics can be used to measure how quick a system can adapt its energy usage to changes in application characteristics.
