Randomized Computation Offloading Algorithms for Mobile Cloud Computing

Abstract

Computation offloading occurs when a mobile application arranges for tasks to be executed remotely, rather than running them locally on the device itself. This mechanism can be used to reduce mobile device energy consumption, resulting in improved battery lifetime. In this thesis, new algorithms that generate offloading decisions for mobile device applications are presented. Randomization is the main technique that is explored, where the algorithms iteratively improve an offloading decision vector by generating random bit strings that represent the task offload decisions. If fragments of these bit strings improve the performance, they are incorporated into the decision vector in a process similar to genetic optimization. In the experiments reported in this thesis, the proposed algorithm typically find good to excellent quality solutions, with low computational overhead compared to existing algorithms. Furthermore, the algorithms scale gracefully as the problem size grows, maintaining high computational efficiency and good solution quality.