Time and space correlation in BGP messages

To quantify the statistical dynamics of the BGP, we analyze the temporal and spatial correlation of macroscopic BGP message flows obtained by passive measurement. We show that the time series for the number of announcement and withdrawal messages has little correlation in time, unlike the statistical behavior of traffic volumes. This indicates that there is little possibility of a cascading failure, in which a failure causes following failures, and that the occurrence of burst of BGP messages has a Poisson nature. We also point out that there is space correlation with the delay between the flows for the different measurement points. Namely, even from macroscopic and passive measurement, we show that the propagation delay of routing information from one measurement point to another point can be statistically estimated.