Internet indirection infrastructure (i3)

The basic idea in this paper is as follows. Instead of a sender directly transmitting packets to a receiver, the receiver first indicates through a trigger what data id it is interested in and where to send it to, while a sender simply sends data conceptually to a data id. In practice, the sender first looks up which server is responsible for handling triggers corresponding to a data id by using some DHT (Chord in this paper) and sends it to that server. An extension is where the data id and receiver addresses can be stacked in order to provide other functionality such as third party processing of data (similar to the outsourced firework?). All these takes place on an overlay network.

Naturally, a key question is how performance is impacted by such an arrangement. Simulations show that latency goes up by about 1.5-4 times, depending on the number of samples used in a heuristic to find an identifier with a closest server. This is for the case when the sender already knows which forwarding server to use. For the first packet, where the sender needs to look up the server, the latency can be up to 15 times larger.

As before, who participates in the DHT for trigger lookup? All the folks who wish to use i3, or a dedicated set of servers? I suppose to figure out what id to use for a certain content, users would have to use some out-of-band communications.