Abstract

The Internet is changing rapidly. One of the consequences of this change is a growing need for higher quality of service. A corollary of the need for higher quality of service is the need for accurate and extensive measurement, including the need to measure bandwidth. Currently the most accurate bandwidth measurement techniques are to directly measure the fastest rate that traffic can be sent through a network. Wide-scale deployment of these `heavy-weight' bandwidth tests can overwhelm the network with test traffic. Accurate measurement of bandwidth is difficult if simple large data volume techniques are not used but there is some current research in this area. We describe existing techniques that attempt to estimate the bandwidth of both links and paths while attempting to use as small a quantity of data as possible. These techniques must operate from only the end points of a connection, and must not require specialist software be deployed into the core of the network. We explain the theory behind two popular bandwidth estimation techniques, single packet and packet pair techniques. The discussion includes the goals of the techniques and the problems each encounters. We also describe some of the improvements that have been suggested. Finally, we explain the differences between current techniques and the goals we are trying to achieve and give an early proposal describing the areas we plan to develop further.

Citations