Using loss pairs to discover network properties
Jun Liu, Mark E. Crovella · 2001
Accurate estimation of network characteristics based on endpoint measurements is a challenging and important problem. In this paper we propose and evaluate a new tool for the discovery of network charactertics called loss pairs. A loss pair is a pair of packets travelling the same path and close together in time, such that exactly one of the two packets is dropped in the network. Loss pairs provide insight into network conditions close to the time when loss is occuring. Using loss pairs, we show that it is possible to characterize the packet dropping behavior of drop-tail and AQM routers internal to the network, from measurements made at end systems. Furthermore, we show that loss pairs can be used to estimate bottleneck buffer sizes of droptail routers. An attractive feature of loss pairs is that they lend themselves naturally to use in a passive-measurement setting. Using such an approach, we show that loss pairs provide estimates that are generally accurate and robust over a wide range of network conditions.