Quantitative Analysis of the Correlation of Round-Trip Times between Network Nodes

Round-trip time (RTT) is defined as the round-trip delay of a data packet between a source and a destination node and is a very important metric for measuring the performance of network applications. The measurement and estimation of RTTs has received a great deal of attention in network measurement research over the years. An interesting question is how RTTs from different source nodes to a common destination node are correlated, which describes to what extent the RTTs exhibit the same changing trend. Such correlation is useful since it would make it feasible to only make some nodes perform RTT measurement while some others estimating RTTs based on the correlation, thus reducing the overhead incurred by RTT measurement. In this paper, we perform a quantitative analysis on the correlation of RTTs, in particular, how network topological properties can affect such correlation. Based on experiment data, our analysis shows that (1) correlations are not much different for path length combinations that only differ in the length of the shorter private path but will become the weakest in statistics when the lengths of the two private paths become the same; (2) correlations are not much different for path length combinations that have the same path length ratio (which is defined as the ratio of the length of the common path to that of the longer private path); and (3) correlation increases along with path length ratio with ratio 1/1 being the inflection point. The results from our analysis of RTT correlation can be applied not only to RTT measurement and estimation, but also to the deployment or selection of measurement nodes in a general network measurement infrastructure.