Passive Access Capacity Estimation through the Analysis of Packet Bursts

Downlink capacity is the most advertised quality parameter of broadband Internet access services, as it significantly influences the user perception of performance. This paper presents an automatic computation method of such a capacity from a measurement point located inside the network. The method is fully passive as it takes advantage of existing TCP connections. It does not inject additional traffic in the network and does not require end-host collaboration. The method takes advantage of the bursty nature of TCP to apply the packet-dispersion approach to TCP segment sequences (packet trains) rather than to segment pairs. This results in a sensible reduction of noise impact on rate estimation. We present an analysis of the effects of the interfering traffic in the access link on rate estimation. We show that it is possible to detect and drop TCP packet trains affected by interfering traffic and to identify and process the packet trains that are not affected by interfering traffic. The proposed method has been validated by means of a set of experiments on ADSL and fibre Internet access services, which are described in the paper. Applications of the proposed method are i) to provide a passive SLA verification method to Internet Service Providers toward Access Service Providers, ii) to support widespread Internet access capacity measurement campaigns, and iii) to perform constant monitoring of access links for fault detection.