Centralized versus distributed replication model for multicast replication

Abstract

In 1990, Deering proposed IP multicast - an extension to the IP unicast service model for efficient multipoint communication. The multicast service model offered two key benefits: the efficient use of bandwidth for multipoint communication and the indirection of a group address which allows for network-level rendezvous and service discovery. This work describes a statistical analysis model that was used to compare the bandwidth used for distributed versus centralized multicast replication. The analysis uses probability to determine the amount of bandwidth that would be needed to serve a population of subscribers using a broadcast TV service. Because the maximum expected use of the broadcast service as a percentage of video subscribers (broadcast peak take rate) is typically much higher than that for the VoD service, the amount of bandwidth that is needed for broadcast and VoD services can be much higher when the broadcast service is delivered as unicast streams rather than by means of distributed replication.