Cloudlet-Assisted Heuristic Offloading for Mobile Interactive Applications

Abstract

Offloading the computation-intensive task of mobileinteractive applications into cloudlets has become a popularway to reduce the response time in mobile cloud computingarea. In response to the rapid change of mobile environment, researchers have proposed online offloading decisions for variousapplications. However, the pruning method they used wouldperform poorly for the non-linear topology of applications. Inthis paper, we propose a heuristic to search out the near-optimaloffloading scheme for the non-linear topology under the real-timeconstraint. In order to reduce the searching space, wefirst leverage Partial Stochastic Path (PSP) analysis to randomlydivide the topological graph into a set of linear paths. Thenfor each linear path, One-climb Policy is applied to obtainthe optimal offloading decision as there is at most one-timeoffloading to the cloudlet if ever. In addition, among the multiplerandom divisions, the one with the most superior performance isselected as the final division. Lastly, we can get the approximateoptimal solution with the minimum completion time in real-time. Simulation results illustrate that our heuristic exhibits anexcellent performance for non-linear topology and outperformsthe state-of-the-art approaches.