Paper review: Quality Adaptation for Congestion controlled Video Playback over the Internet

Reviewer: Kevin Hofstra

  1. Video playback over the web must not only deal with varying bitrates for different customers but also with fluctuations in bandwidth caused to congestion.  What is the best way for having different quality video for the different bandwidth restrictions of users?  How can the streams be made to avoid stoppages and degeneration in the event of network congestion?
  2. This paper presents a mechanism for using buffered layers of video to present a fluid stream to many users of differing bandwidths at different qualities.  They have a study of how many layers should be present, how much buffer they should receive and the Kmax that will provide the most smoothing.

A.  The video will be sent in distinct layers with each layer adding quality when available, and all streams containing at least the base layer.  This allows each user to receive the stream with those that can handle higher receiving better quality.  This is also important in the case where a high bandwidth user experiences high congestion, and instead of losing his entire high bandwidth stream, he loses the added quality layers, while retaining the base stream.

B.     It is possible to calculate the Optimal Inter-Layer Buffer Allocation.  Efficiency is gained by trying to prevent the loss of useful buffer when a layer is lost, but still making sure that each layer is able to buffer enough that it may still be kept for playback.

C.     The Kmax factor may contribute to smoothing.  This ensures that a layer is added if and only if it has the availability of optimal buffer allocation for recovery from Kmax backoffs.  We can then prevent the continuous adding and removing of a layer that is not semi-permanent.

  1. Critique the main contribution
  2. System researchers and builders should recognize that streaming video is distributed over many varying bandwidths, and each also has the possibility of heavy congestion.  In order to ensure a constant stream, the most effective method is always having a base stream and adding layers of additional quality when bandwidth or congestion allow.  Further areas of research that will be needed include the optimal time for additional layers, and more research into optimizing the buffer across the layers depending on probability of losing a layer.