Optimal Delay for Media-on-Demand with Pre-loading and Pre-buffering

Broadcasting popular media to clients is the ultimate scalable solution for media-ondemand. The simple solution of downloading and viewing the medium from one channel cannot guarantee a reasonable startup delay for viewing with no interruptions. Two known techniques to reduce the delay are pre-loading and pre-buffering. In the former an initial segment of the medium is already in the client buffer, and in the latter segments of the medium are not transmitted in sequence and clients may pre-buffer later segments of the medium before viewing them. In both techniques, clients should be capable to receive streams from channels at the same time of handling their own buffer and view the medium from either one of the channels or the buffer. This paper considers broadcasting schemes that combine pre-loading and pre-buffering. We present a complete tradeoff between (i) the size of the pre-loading; (ii) the maximum delay for an uninterrupted playback; (iii) the number of media; and (iv) the number of channels allocated per one medium. For a given B the size of the pre-loading as a fraction of the medium length, for m media, and for h channels per medium, we first establish a lower bound for the maximum delay, D, as a fraction of the medium length, for an uninterrupted playback of any medium out of the m media. We then present an upper bound that approaches this lower bound when each medium can be fragmented into many segments. ∗The research was partially funded by NSF Grant No. CCF-0223578