On Data-driven Network Performance Modeling for Mobile Cloud Computing

Computationally intensive mobile apps may be migrated to a cloud infrastructure for faster remote execution. Decreased execution time and lower energy consumption at the mobile device are the expected benefits when offloading the application to the cloud. The migration decision can be taken based on a continuous-time Markov model that considers network quality, cloud and mobile device capabilities, as well as migration costs, as we have shown in previous work. One of the influencing dynamic characteristics is the network performance. In this work, we focus on characterizing network performance under node mobility in terms of throughput and latency. Our final goal is to derive a mobile performance model that goes beyond an on-off network model. The analysis is based on performance measurements taken on a train while commuting. By clustering the measurement data, we derive a realistic network model.