| In this thesis, available bandwidth estimation tools including Delphi, Spruce, Pathchirp, and Pathload are evaluated. Each method has disadvantages in satisfying the requirements of fast, accurate and non-intrusive. Among all, Pathload is the most reliable tool that considers the IC/CS phenomenon. Pathload sends multiple trains called fleets. Each train consists of 1200 same-size packets, with the same inter-packet time gaps. After each fleet is sent, statistical evaluation of the inter-packet time gaps at the destination compared to the inter-packet time gaps at the sender determines whether the trend of the inter-packet time gaps in that fleet has been increasing, non-increasing, or undeterminable. The transmission rate for the next fleet is adjusted based on the trend of the current fleet until the results converge. One challenge of Pathload is that the underlying one way delay detection (OWD) and the method of handling IC/CS effect is far from perfect. In this thesis, the previously proposed algorithm for efficient OWD trend detection based on a statistical method is implemented into the Pathload. Then, the enhanced Pathload is compared to Pathload. Experiments in our laboratory network and across several Internet paths clearly show that our enhanced Pathload significantly outperforms tests using Pathload with its original OWD trend detection algorithm. |
