Characterising and exploiting workloads of highly interactive video-on-demand

Abstract

This paper presents a detailed characterisation of user behaviour for a series of interactive video experiments over a 12 month period, in which we served popular sporting and musical content. In addition to generic VCR-like features, our custom-built Video-on-Demand application provides advanced interactivity features such as bookmarking. The dramatic impact of such functionality on how users consume content is studied and analysed. We discuss in detail how this user behaviour can be exploited by content distributors to improve user experience. Specifically, we study how simple dynamic bookmark placement and interactivity-aware content pre-fetching and replication can reduce the impact of highly interactive media on CDN performance.

Extracting viewer interests for automated bookmarking in video-on-demand services

“…Furthermore, by carefully studying the figure, we find that the distribution curve after 5 seconds, which cover from median to long inter-seek times, actually Table 1 and from all videos decays slower than power-law distribution. The observation suggests that it would be impropriate to model the inter-seek times using a single power-law distribution, like lognormal [4] or Weibull [14] as suggested in previous studies.…”

“…While in [14], Garcia et al proposed to use two exponential distributions to estimate a viewer's total watching length in a video, and how the two distributions are mixed depends on the viewer's interest level; they also reported that viewers' inter-seek times can be modeled with a Weibull distribution. Brampton et al [4] suggested that the inter-seek times follow a lognormal distribution, and through a controlled experiment, they showed that with manually positioned bookmarks, viewer will have a much better service experience. Our work differs from these studies in that we propose to use mixtures of two lognormal distributions to fit the empirical inter-seek times, moreover, we present a bookmarking algorithm that accurately positions bookmarks on various types of videos with high accuracies.…”

“…3. For comparison, we also apply the models of one lognormal distribution as suggested in [4] and one Weibull distribution in [14] to fit the empirical data. For each fitting, we compute the determination coefficient R 2 , which indicates how good the model fits the empirical data, and display the values in the figures.…”

“…In a bookmark-assisted VoD service, a viewer could choose to directly jump to a bookmark, and the client could prefetch all the bookmarked contents in advance for reducing the seeking delay. A recent study [4] reported that when accurately positioned bookmarks are available, viewers' VCR experiences can be improved considerably.…”

Enriching Viewer Options in P2P Media Streaming Applications

“…For instance, the streaming clients do not accept customized choices of experienced viewers, such as discrete segments Viewer Provider Content Time segment of the original media content. Brampton et al [31] make progress in this direction by offering bookmark-based access to an on-demand media streaming service, which in turn results in considerably different patterns of the viewer behavior.…”