MAsCOT: Self-adaptive opportunistic offloading for cloud-enabled smart mobile applications with probabilistic graphical models at runtime

Although extensive progress has been made in Mobile Cloud Augmentation, automated decision support on the device that enables the opportunistic and intelligent use of cloud resources is missing. Furthermore, we need solutions with reflective capabilities that can handle a changing environment and runtime variability. To simplify the deployment of smart mobile applications, we present a framework with retrospective decision support based on reinforcement learning to cater for various resource-performance trade-offs. We have adopted the MAPE-K (Monitor-Analyse-Plan-Execute-Knowledge) control loop architecture and realized the loop with Dynamic Decision Networks to manage self-adaptation at runtime. Our experiments show that our framework is capable of intelligently inferring appropriate decisions with an acceptable performance overhead of 10 milliseconds on mobile devices