End-to-end inference of link level queueing delay statistics