Inferring arrival rate statistics of individual flows from aggregated-flow rate measurements
Masato Tsuru, Tetsuya Takine, Yuji Oie · 2003
For a statistical perspective of global traffic flows on a large network, the authors have proposed a method for inferring unobservable arrival rate statistics of individual flows based solely on measurements of aggregated-flow rates, which can be easily measured at network links such as router interfaces. The above method, while having potential for inferring arrival rate statistics of various kinds of flows without the need to identify the flow to which each packet belongs, originally had limited practical applicability. Therefore, to extend the range of application of the method, the following improvements are proposed: i) finding the minimum effective rate of each flow and inferring a statistic of the difference between minimum effective rate and the flow rate itself, ii) quantizing rate values into bins with an appropriate bin size; and iii) inferring rate statistics of flows indirectly from those of other "well inferable" flows. Simulations are performed, and the proposed improvements are shown to allow the mean rate inference of each flow in cases that previous methods have been unable to treat.