Deep Reinforcement Learning Based Computation Offloading for Not Only Stack Architecture

New technologies emerge with regard to computation offloading, allocating parts of applications to powerful servers to meet demands for the massive traffic of mobile devices. In this paper, we study an offloading framework enabled by Not Only Stack (NO Stack). NO Stack, a promising architecture adopts the virtualization technology, guarantees the implementation of machine learning. A deep reinforcement algorithm triggers energy-efficient strategies with handover control when a mobile device moves among cells. Offloading decisions are formulated by an agent with the feedback from the complex environment. Simulation results demonstrate that our proposed offloading system based on NO Stack reduces the energy consumption and cuts down handover rates for mobile devices, compared with the conventional system when executing a service workflow.