Improving the scalability of an application-level group communication protocol

We analyze the scalability of a control protocol providing a user-level group communication service. This control protocol, called HBM (host based multicast), can be used when native multicast routing is not available. HBM is, by nature, centralized, everything being controlled by a single host, called the rendezvous point, or RP. This feature naturally leads to scalability issues; however, we show that scalability can very easily be improved. In particular, the control information being exchanged between the group members and the RP can be rate limited not to exceed a given threshold. This is made possible by a careful modeling of the control traffic exchanges and is validated by many experiments carried out with our prototype.