Power and Performance Optimization of Smartphones

Abstract

Smartphones are often used for prolonged periods while browsing over internet, video/audio streaming, gaming etc., without being docked. Their high performance along with minimal power consumption is a challenge to be met. Intel's rst smartphone launched recently was its rst step into the smartphone world. This was launched with a single core and 1.6 GHz maximum CPU frequency. The roadmap in the coming years would have dual core and quad core platforms that are in their preliminary stages yet. This thesis report discusses a few of the various challenges to increase the performance of these devices in order to be competitive. Starting with an overview of Intel's Atom processor architecture and its CPU power states, we take a look at the software stack of the android OS in these devices. The report discusses two browsing benchmarks and a graphics benchmark in order to investigate on how would they utilize the upcoming platforms with additional cores from performance, power and energy e ciency point of view. A workload performance should scale with additional cores and the active power state should be divided among the cores, thus resulting in power savings. Also the watts per performance unit should decrease for the workload to be energy-e cient. An investigation of the typical "Day Use" cases and the utilization of additional cores for these scenarios is done. This allows one to predict the number of cores required to handle these use cases or workloads. The report also discusses single-tasking verses multi-tasking on dual core x86 platform. This shows the distribution of the CPU utilization among various CPU frequencies when additional cores are available. As a case study, the thesis talks about android browser followed by a page-load analysis on an x86 platform compared to an ARM device. The end goal of this case study is to identify both software and hardware level bottlenecks by using number of debugging techniques in order to optimize the page load time of web pages in android browser on x86 platform.