Abstract

In this short paper we briefly describe the results of analyzing a large-scale data set of actual YouTube stall patterns collected world-wide as part of Columbia University’s YouSlow project, and how we used it to create a simple model for generating realistic stalling patterns with a given number of stalls, a given average stall duration, and a pattern structure (describing the relative length of the stalling events). These stall patterns can be used to perform subjective assessment of HTTP video under realistic conditions. A tool for generating the patterns and the data accompanying this paper has been released for the research community to use.