---

Observed performance (i.e. delivered quality) for a spectrum of adaptive video streaming applications primarily depends on their obtained bandwidth over the Internet. Because of the "streaming" nature of delivery in these applications, not only average obtained bandwidth but also its variations over short timescales (e.g. several seconds) could significantly affect delivered quality. More importantly, as the degree of interactivity in these applications increases, their ability to use buffering to absorb bandwidth variations decreases, and thus they become more sensitive to variations in obtained bandwidth over shorter timescales. For example, video playback applications, lecture-mode streaming of live video and point-to-point video conferencing are able to buffer minutes, seconds and hundreds of milliseconds worth of content, respectively. Therefore, video conferencing applications are more sensitive (and must react) to variations of bandwidth over much shorter timescales compare to video playback applications. These applications can significantly benefit from characterization of obtained bandwidth over their corresponding timescales to effectively cope with bandwidth variations while minimizing its impact on delivered quality. For instance, having a priori knowledge about duration and degree of short-term variations in bandwidth on a target path enables quality adaptation mechanisms to distinguish transient changes from major shift in bandwidth

In this project, we conduct measurement-based characterization of congestion controlled bandwidth over short timescales. Since Internet applications should be congestion controlled, we believe that such characterization should be conducted for congestion controlled bandwidth. More importantly, characterization of congestion controlled bandwidth should be performed over different timescales. We are investigating how characterization of congestion controlled bandwidth change across different paths to verify whether variations of bandwidth for each path exhibits a statistically unique behavior or there are several classes of behavior that are common among different paths.