ABSTRACT:

The end-to-end available bandwidth of an Internet path is a desirable information that can be exploited to optimize system performance. Several tools have been proposed in the past to estimate it. However, existing measurement techniques were not designed for large-scale deployments. In this paper we show that current tools do not properly work where multiple probing processes share a portion of a path. We provide experimental evidence to quantify the impact of mutual interference between measurements. We further analyze the characteristics of popular tools, quantifying (i) the impact of mutual interference, (ii) the total overhead imposed to the network and (iii) the intrusiveness of the measurement process in a large-scale scenario. Our goal is to effectively quantify the impact of concurrent measurements on current estimation techniques and to offer some simple guidelines for dimensioning a large-scale measurement system.