What is really needed is a general architecture for all internet distribution. This architecture should have mechanisms for both space-based and time-based multicast. It should be able to seamlessly incorporate multicast media (LANs, satellite) and IP multicast where they exist and where their use is appropriate. It should have an easy evolutionary path, by which I mainly mean that it shouldn't require changes to existing infrastructure, including host operating systems. It should have a fully scalable naming scheme. And it shouldn't suffer from any of the nagging problems of IP multicast.
This paper presents an architecture that holds promise as the general architecture for all internet distribution. I call it yoid, for reasons that will be clear shortly.
Simply put, yoid is a suite of protocols that allows all of the replication and forwarding required for distribution for a given application to be done in the endhosts that are running the application itself. In other words, yoid works in the case where the only replicators/forwarders (distributors) of content are the consumers of the content themselves. Let me be clear. Yoid does not force all distribution to be done by the content consumers--it can also be done by ``servers'' in the infrastructure. Nor does yoid prevent the use of IP multicast--it is used, but only where it exists and where its use is appropriate. With yoid, neither infrastructure servers nor IP multicast are necessary for distribution, but they can be brought to bear where economics or application requirements make them appropriate. When IP multicast is used, it will generally be confined to local networks. Islands of locally scoped IP multicast groups will be connected by global yoid.