Analysis of local re-execution in mobile offloading system

Richer functional mobile applications are developed through offloading the heavy computation from resource constrained mobile devices to powerful cloud servers. For guaranteeing the completion of such tasks, strong connectivity with the servers is essential. However, the wireless network is not sufficiently reliable. It is often unable to satisfy the offloading condition, and thus interrupts the program execution. To deal with this problem, locally re-executing the offloaded computation in the mobile client is an efficient way to maintain the continuity of the application. In this paper, we present a synthetical method to find the optimal moment for launching the local re-execution. First, we develop a general program engine to implement offloading applications. Then we design a Stochastic Activity Network (SAN) model based on the engine structure and three metrics to evaluate the engine's performance. The model parameters are set to values obtained from experimental data of our engine. We compare the performance of different local re-execution launching intervals with the model simulation and find the optimum. Last, the optimal launching interval is returned to configure the engine and its efficiency is verified with experiments. The result shows that starting the local re-execution after an appropriate interval provides a better performance than always offloading, and our model can effectively find the optimal launching interval.