The Trade - o s of Multicast Trees and Algorithms Liming

Multicast trees can be shared across sources (shared trees) or may be source-speciic (shortest path trees). Inspired by recent interests in using shared trees for interdomain mul-ticasting, we investigate the trade-oos among shared tree types and source speciic shortest path trees, by comparing performance over both individual multicast group and the whole network. The performance is evaluated in terms of path length, link cost, and traac concentration. We present simulation results over a real network as well as random networks under diierent circumstances. One practically signiicant conclusion is that member-or sender-centered trees have good delay and cost properties on average , but they exhibit heavier traac concentration which makes them inappropriate as the universal form of trees for all types of applications. 1 Introduction Multimedia communication is often multi-point and has contributed to the demand for multicast support. moreover, multimedia applications are often high bandwidth and delay sensitive. Therefore it is essential to consider the eeciency and quality of multicast distribution trees. The problem of computing the optimal multicast path, in the shape of a tree or a group of trees, has many potential solutions. Traditional IP multicast protocols have solely used source rooted shortest path trees, not because it is the best bandwidth-saving strategy, but based on the fact that today's multicast applications are primarily small scale and local area. Factors such as protocol simplicity and convenience of development have dominated protocol design activities. In the context of large-scale, wide area networking, where resources are not as plentiful as in the local area network, the previously-ignored diierences between tree types may lead to signiicant diier-ences in cost and performance of multicast services. However , to date there have not been systematic comparisons among the diierent solutions. In this paper, we judge the quality of a tree according to the following three dimensions: