Uloof user-level online offloading framework: user-level online offloading framework

Abstract

Our fast paced day-to-day life has been facilitated by the use of mobile phones. Our growing demand on applications that are used for business, entertainment and research areas has been pushing forward the development of the mobile platform. Problems arise when such devices intrinsically furnished with limited resources (e.g. battery, memory and CPU) can not offer the expected quality of service and can barely go through a normal working day without a recharge. Cloud-assisted Mobile Offloading consists of techniques to save battery and opportunistically improve execution time of applications. In this scenario, code is optimally executed on the cloud, depending on some heuristics, culminating in lower mobile resource usage. Offloading frameworks provide an easy start-up and a powerful set of tools to enable Offloading in mobile applications. Identifying common shortcomings of previous works, this thesis develops ULOOF, an User Level Online Offloading Framework. Our work is a general-use transparent framework that was designed to work in a plug-and-play fashion. We employ an accurate online energy model derived from real device data; and a time model that employs a black-box approach for execution time estimation. Our contribution also includes prediction of latency according to user location, resulting in enhanced contextualized Offloading decisions while incurring low overhead. We created two experimental applications, and their evaluation indicate that ULOOF can reduce the application's energy usage up to 80% and lower execution time up to 47.64% under our setup.