Maximising user experience through intelligent resource management in mobile systems

Abstract

Mobile devices are required to deliver increasing performance and high quality of experience (QoE) to users, despite comparatively slow advances in battery technology. While the complexity of modern mobile devices has increased exponentially, both in terms of computing capabilities and applications with diverse requirements, achieving battery life goal is crucial if the users’ QoE is to be maximised. Few studies have considered longer-term resource budgeting, but they either fall short of users’ battery life expectations or an unnecessarily over-constrained QoE. This research aims at maximising the QoE across the duration of battery discharge for a user and their requirements.To achieve the stated aim, this thesis first explores how to predict plug-in times and energy demand, along with how the compute performance of a mobile device can be adapted to balance QoE and battery life requirements. This leads to the proposal of QUality of experience Adaptive REsource Management (QUAREM), an adaptive Dynamic Voltage and Frequency Scaling (DVFS) approach for maximising QoE. QUAREM learns and predicts users’ plug-in times and energy demand and exploits the DVFS via online monitoring and adaptive frequency setting to achieve the battery life. Evaluationshows an average QoE improvement while meeting daily battery life expectations.Next, the thesis explores the relationship between display brightness and how it affects user QoE, as well as how it can be adapted to further tradeoff battery life and QoE. aCADS, an adaptive brightness scaling approach that leverages insights from user perceptions of content and ambient light variations to maximise QoE, is proposed. This is achieved through the learning and classification of each sample into predefined content and ambient light clusters and the adaptive scaling of the display brightness using an energy model. Compared to state-of-the-art, aCADS improves QoE by up to 32.5 %.To effectively manage and balance battery life and QoE while accommodating diverse usage patterns beyond processing elements’ (PEs) or display subsystem’s capabilities, an adaptive brightness scaling approach along with PE’s DVFS is proposed. This is achieved with an efficient workload- and context-clustering approach and an appropriate adaptive resource evaluation and allocation. Evaluation on smartphone shows QoE and battery life improvements of 9% and 35 %, respectively, over state-of-the-art.

More information

Identifiers

ORCID for Samuel Isuwa: orcid.org/0000-0002-2235-4091

ORCID for Geoff Merrett: orcid.org/0000-0003-4980-3894

Catalogue record

Export record