Bandwidth estimation is a very important topic, especially in wireless Networks, to determine the data rate available on a network path or link, usually expressed in bits per second. Having information available can help to develop better mechanisms for e.g. gateway selection, channel selection, routing, etc. Several approaches can be distinguished that allow to determine the available capacity.
In those approaches, researchers try to model the network using e.g. Markov processes and use those models to predict e.g. the throughput. As an example, the Bianchi model has been extensively used to predict throughput of 802.11 based WLAN networks. Several extensions have been developed to model e.g. the achievable throughput in an 802.11 based Ad-Hoc network using the Bianchi Model. However, analytical models for networks are highly topology-dependent and may not be able to provide useful information for specific scenarios.
In those approaches, researchers try to apply active measurement techniques to estimate available bandwidth or capacity of a link or a path. Several measurement tools have been developed, such as Spruce, TOPP, Pathchirp, IGI or Pathload, mostly based on sending probing packets which can create significant overhead and severely interact with the real data currently transmitted in the network.
Here, available bandwidth is inferred based on calculations taking into account several parameters such as estimating the channel capacity of a link using e.g. channel busy ratio. However, several challenges arise when such techniques should be applied in a multi-hop scenario due to collissions and different view of the channel to come to a good estimate of the per link or end-to-end estimate.
As a result, estimation of available link and path capacity is a very relevant research topics and especially in a wireless (and multi-hop) context requires further research efforts.