Scheduling content multicast trees with minimal repetitions

Abstract

This paper discusses an approach for distributing a large number of multimedia (video) titles to users located at the leaves of a tree network. The proposed approach is to build multicast trees, over which the content files are sent to user disks, followed by delayed local playback of the content. The work concentrates on exact and heuristic methods for an efficient tree packing in several download periods. In order to satisfy all the requests accumulated in a certain time period, we choose to minimize the number of times a multicast tree is split because of capacity constrains. We show that this objective provides lowest link utilization and good convergence both in the integer formulation and in the developed heuristics (of type tabu search, iterated local search, and variable neighborhood search). The model can be also used in a planning setting to answer the question on the maximum user request rate that can be supported by a certain network topology.