It is known that unless NP ⊂ DTIME(n log log n ), no polynomial-time approximation algorithm for the multicast problem can have approximation ratio less than (log n) in n-node digraphs under the edge-disjoint paths mode of the line model. In this note, we give a polynomial-time O(( min + log n)=(log( min + log n)))-approximation algorithm, where min is the smallest integer k such that there exists a rooted directed tree of maximum out-degree k, spanning the considered digraph.
Statement of the problemGiven a source-node s in a network, and a set D of destination-nodes, multicasting from s to D consists in transmitting a piece of information from s to all nodes in D using the communication facilities of the network. Broadcasting is a particular case of multicasting in which the destination set consists of all nodes of the network. Multicasting and broadcasting are two of the basic operations upon which network applications are frequently based nowadays. They hence gave rise to a vast literature, covering both applied and fundamental aspects of the problem (cf. [8,22] and [19], respectively).As far as graph-theoretic aspects are concerned, the multicast problem can be expressed as follows. We are given a graph G = (V; E), a node s ∈ V , and a set D ⊆ V . We are looking for the most e cient multicast protocol from s to D in G. The nature of the "protocol" and the measure of its "e ciency" depend on the communication model.The local model assumes that transmissions proceed by synchronous calls between the nodes of the graph. It is assumed, moreover, that (1) a call involves exactly two neighboring nodes (locality constraint), (2) a node can participate in at most one call at a time (single-port constraint), and (3) the duration of a call is 1 (atomic constraint). A multicast protocol is then described by the list of calls placed between the nodes of the graphs. The e ciency of the protocol is measured in terms of number of rounds, where round t is deÿned as the set of all calls performed between time t − 1 and time t, t = 1; 2; : : : . This model has been intensively investigated, for both speciÿc and arbitrary topologies (cf. [16,20] and [1,2,[9][10][11]21,23,24], respectively).The line model relaxes the locality constraint, and allows calls to be placed between non-neighboring nodes. A call is then a path in the graph, whose two extremities are the "caller" and the "callee". (Note that, as opposed to the models in [4,5], the intermediate nodes between the caller and the callee do not receive the information which just "cuts through" the routers attached to them.) Several modern technologies (e.g. single-hop WDM for optical networks) support transmissions whose costs are distance-invariant, and are hence good applications for the line model.Two main variants of the line model have been investigated.E-mail address: pierre@lri.fr (P. Fraigniaud).