User-centric optimization techniques for smartphones

Abstract

Recently, smartphones have become an integral part of everyday life. In addition, as smartphone users are expecting their devices to deliver PC-level user experience, numerous design requirements are rapidly emerging with the technology development. In order to meet the demanding system design requirements, many conventional techniques, whose basic concepts are almost same as those of the traditional computing devices such as PCs, are applied to the smartphones. However, as truly personalized and interaction-oriented devices, the smartphones have distinct characteristics which distinguish the devices from traditional computing devices. Therefore, it is highly required to understand and to analyze the distinctive inherent characteristics of smartphones for a provision of a new novel opportunity for system optimization.

In this dissertation, we propose new user-centric optimization techniques to satisfy various design requirements of smartphones such as energy efficiency, effective thermal management and rapid responsiveness without any performance degradation by taking advantage of high-level information from the smartphone users.

We first introduce a new definition of the response time, the user-perceived response time, which is known to be a critical metric for the quality of user experience of the smartphone. We also present a user-perceived response time analyzer for Android-based smartphones, which can identify the user-perceived response time of smartphone apps during run time. Based on on-line identification of the user-perceived response time, we propose a novel CPU power management framework, which enables more aggressive low-power techniques to be employed while executing display-insensitive parts of task executions. Our experimental results on a smartphone development board show that the proposed technique can reduce the CPU energy consumption by up to 65.6% over the Android's default ondemand cpufreq governor.

Second, we propose a novel dynamic thermal management (DTM) technique for smartphones, which ensures the quality of user experience during the execution of display-sensitive parts without any thermal violations. In the proposed DTM technique, in order to identify that the current execution could affect the visible portion of the display, we develop a user-perceived response time prediction model for each interactive session based on statistical analysis of the user-perceived response times for the past interactive sessions. By exploiting the on-line prediction of the user-perceived response time, the proposed DTM technique carefully makes the DTM decisions for a higher quality of user experience. Our experimental results on an ODROID-XU+E board show that the proposed technique can improve the user-perceived performance by up to 37.96% over the Androids default DTM policy.

Third, we present a personalized optimization framework for smartphones which can provide valuable high-level hints for optimizing the smartphone design requirements. The main goal of the proposed framework is collecting an app usage log of a smartphone user and analyzing the collected log so that particular usage patterns, if any, can be effectively identified. In order to identify app usage patterns, a couple of app usage models are also proposed. Based on the app usage models developed, we also propose a launching experience optimization which avoids unnecessary app restarts considering the detrimental effects of the restart on user experience from the perspective of performance, energy, and loss of previous state. Our experimental results on the Nexus S Android reference phones show that our proposed optimization technique can avoid unnecessary application restarts by up to 78.4% over the default LRU-based policy of the Android platform.

Based on the evaluation for each technique, we verified that the user-centric optimization techniques improve the quality of user experience in terms of energy efficiency, effective thermal management and rapid responsiveness over previous system-centric techniques.