Decision Support for Computational Offloading by Probing Unknown Services

Mobile Cloud Computing (MCC) leverages resourceful data centers that are distant (aka the cloud) or closely located (aka edge servers) for computational offloading to overcome resource limitations of modern mobile systems like smartphones or IoT devices. Many research works investigate context-aware offloading decision algorithms aiming to find the best offloading system at runtime. However, all approaches require prior knowledge of the offloading systems or a running service profiler on the backend system. In this paper, we present a novel approach that overcomes this issue by first probing available unknown services such as nearby cloudlets or the distant cloud, and networks in an energy-efficient way at runtime to make better offloading decisions. For that, we investigate a probing strategy to assess these unknown services by offloading micro tasks and accurately predicting the performance for larger offloading tasks using regression models. Our evaluation on three algorithms with different time complexities shows that we achieve high prediction accuracies up to 85.5%, already after probing of two micro tasks running in the range of few milliseconds. To the best of our knowledge, this is the first supplement approach for offloading decision support that can handle unknown third-party services requiring no prior knowledge about these offloading systems and making no assumptions for real-world deployments.