The importance of granularity in multiobjective optimization of mobile cloud hybrid applications

Mobile devices can now support a wide range of applications, many of which demand high computational power. Backed by the virtually unbounded resources of cloud computing, today's mobile cloud (MC) computing can meet the demands of even the most computationally and resource-intensive applications. However, many existing MC hybrid applications are inefficient in terms of achieving objectives like minimizing battery power consumption and network bandwidth usage, which form a trade-off. To counter this problem, we propose a data-driven technique that (1) does instrumentation by allowing class-, method-, and hybrid-level configurations to be applied to the MC hybrid application and (2) measures, at runtime, how well the MC hybrid application meets these two objectives by generating data that are used to optimize the efficiency trade-off. Our experimental evaluation considers two MC hybrid Android-based applications. We modularized them first based on the granularity and the computationally intensive modules of the apps. They are then executed using a simple mobile cloud application framework while measuring the power and bandwidth consumption at runtime. Finally, the outcome is a set of configurations that consists of (1) statistically significant and nondominated configurations in collapsible sets and (2) noncollapsible configurations. The analysis of our results shows that from the measured data, Pareto-efficient configurations, in terms of minimizing the two objectives, of different levels of granularity of the apps can be obtained. Furthermore, the reduction of battery power consumption with the cost of network bandwidth usage, by using this technique, in the two MC hybrid applications was (1) 63.71% less power consumption in joules with the cost of using 1.07 MB of network bandwidth and (2) 34.98% less power consumption in joules with the cost of using 3.73 kB of network bandwidth.