Applying clustering algorithms as core selection methods for multiple core trees

In the current Internet, there are two main types of intra-domain multicast routing protocols: dense mode and sparse mode. Dense mode protocols construct shortest path trees from the sender to the receivers. Sparse mode protocols construct shared trees rooted at a certain router, called core or rendezvous point (RP), which acts as a meeting point between senders and receivers of the same group. While solving important scalability problems in wide area networks, shared trees also bring some drawbacks, the most known of which are traffic concentration and delay. In order to minimize these kinds of drawbacks, a variant of sparse mode using several cores per group emerged. Receivers are attached to the closest core and sources send packets to all the cores with attached receivers of the corresponding group. In this paper three clustering algorithms are presented and compared: max-cut, min-cut and min-concentration. They select the minimal set of cores in multiple core trees that maximizes or minimizes a certain objective function while limiting the QoS degradation according to a certain parameter called clustering ratio. These core selection algorithms are used to show the advantages of using several cores per group over the classical method of using only one.