Predictive context aware mobility handling

The handling of device multi-homing and mobility, such as deciding which network interface to use or when to perform vertical handoff between network interfaces, can be greatly enhanced by considering recent context information. We describe a system for context aware multi-homing and mobility handling which enacts network interface allocation and handoff decisions based on the predicted characteristics of transport layer sockets and network interfaces. Predictions are made using a statistical machine learning technique which can utilise simple context data such as time-of-day and GPS co-ordinates, as well as more complex contextual information such as nearby Bluetooth beacons and internal system state. We present a prototype implementation of the described system and show via experimentation that it enables more timely mobility handling without requiring changes to either applications or to the underlying operating system.