Characterizing network elements and paths using packet loss behavior

This dissertation explores problems that arise in characterizing network elements and network paths. Characterizing network elements and paths can be valuable for network performance improvement, and network simulation and modeling. Three broad kinds of characterization are explored in this dissertation: (a) characterizing buffer space; (b) characterizing packet loss models; and (c) differentiating packet loss types.; In characterizing buffer space of a single network element, two methods are studied: a probabilistic method, and a heuristic method. The probabilistic method characterizes buffer space by making use of stochastic properties of network traffic. The heuristic method makes use of the queuing behavior of the network element. Both methods are evaluated in simulation and are shown to be effective.; In characterizing buffer space and packet loss models of the bottleneck link along an end-to-end network path, this dissertation develops a method called loss pairs. This method is evaluated both in simulation and in experiments conducted over an emulated multi-hop network environment, and shown to be effective given typical assumptions about path properties.; Finally, by making use of the loss pairs method, this dissertation develops a method for end-to-end differentiation of losses in a mixed wired/wireless environment. This method attributes a loss to its most likely cause: congestion or wireless channel fading. Using simulation it is shown that this differentiation method works well, and is superior to a traditional alternative method.