Model-based Decision Framework for Autonomous Application Migration

A dynamically changing state of the run-time environment impacts the performance of networked applications and furthermore influences the user-perceived quality of the applications. By migrating the application between the user's devices that give different user experiences, a good overall user experience can be maintained, without loosing the application session. It is non-trivial for a system to automatically choose the best device for migration. The choice must maximize the user experience quality and take into account that a migration delays the user's work-flow and even may fail. Moreover, the environment state is not directly observable, and needs to be estimated which leads to inaccuracy. We model the automatic migration trigger as a stochastic optimization problem and we propose to use a hidden Markov model combined with a Markov Decision Process (MDP) to solve the problem. The solution generates policies to choose target device for migration that gives the optimal user experience. We analyse these policies in simulation experiments and derive conclusions on which scenarios the model-based approach performs better than a greedy approach, also when considering inaccurate state estimation.