A context-sensitive offloading system using machine-learning classification algorithms with seamless mobility support

Abstract

Mobile Cloud Computing (MCC) enables resource-constrained smartphones to run computation-intensive applications through code/data offloading to resourceful servers. Nevertheless, this technique can be disadvantageous if the offloading decision does not consider contextual information. Another MCC challenge is related to the change of access point during an on-going offloading process, since it impacts on or is impacted by resource scarcity, finite energy, and low connectivity in a wireless environment. This PhD research has developed a context-sensitive offloading system that takes advantage of the machine-learning reasoning techniques and robust profilers to provide offloading decisions with the best levels of accuracy as compared to state-of-the-art solutions. In addition, this work proposes a way to support seamless offloading operations during user mobility through the software-defined networking (SDN) paradigm and remote caching technique to speed up the offloading response time. Firstly, in order to address the offloading decision issue, the approach evaluates the main classifiers under a database comprised of cloud, smartphone, application, and networks parameters. Secondly, it transforms raw context parameters to high-level context information at runtime and evaluates the proposed system under real scenarios, where context information changes from one experiment to another. Under these conditions, system makes correct decisions as well as ensuring performance gains and energy efficiency, achieving decisions with 95% of accuracy. With regards SDN-based mobility support, the results have shown that it is energy efficient, especially considering the low-cost smartphone category, while remote caching proved to be an attractive alternative for reducing the offloading response time.