End-to-end inference of link level queueing delay distribution and variance