Model-based end-to-end available bandwidth inference using queueing analysis