Verifying IP Meters from Sampled Measurements

Traffic measurement and analysis is nowadays a very critical task and requires complex and high-cost traffic analysis equipment and infrastructure. The difficulty increases when high bandwidth connections are to be monitored. Network topology, the network technology used, applications and user behavior influence the overall traffic on the network. For network planning and provisioning is crucial to know the trends of the traffic behavior, that is why the analysis of protocol application is very useful. Most recent low-cost monitoring platforms for high-speed links rely on samples of the whole traffic. The purpose of the measurements range from monitoring and traffic characterization, to charging and billing. It is a must to know the statistical validity of the sampled data and the consequent detailed results obtained by classifying the traffic according to different categories. This fact is even more important when it is intended to be used for charging or cost sharing. In this paper we propose an approach to validate the statistical capture of our system. We compare full traffic measurements which contain all the traffic with sampled traffic measurements, in order to know what is the threshold of a sampled traffic to estimate the real traffic with a certain degree of confidence. Results are presented using traffic from a real network environment and comparing our capture platform results with results derived from the real traffic.