Impact of duration on active video testing

Abstract

There is a growing interest in video performance measurements with emphasis on user experience and several initiatives have been taken to conduct active testing of real video services. A deeper understanding of the variations in media bit rate and its influence on the performance of video playback is needed in order to design better measurements. In this paper, we analyze a dataset of YouTube videos from various genres. We show statistically that most YouTube videos can be represented sufficiently well by the first 1 to 3 minutes of the video. This eliminates the need for running longer tests when network conditions are stable as in the case of fixed networks. We test our observation in an active testing environment that measures video metrics, and recommend based on the results that such tests should run at least for one minute, however, a duration of 3 minutes will help achieve better and more stable results.