Summary: I3

The paper describes a novel protocol for overlay-based Internet
indirection infrastructure that allows applications to implement services
such as multicast, anycast and mobility. This overlay approach that
avoids both technical and deployment challenges inherent in IP-layer
solutions and also the redundancy and lack of synergy in more traditional
application layer based solutions. Like traditional IP-layer based
approaches, sending host no longer needs to the identity of receiving
hosts and location of receiving host need not be fixed. The indirection
approach essentially decouples sending hosts from receiving hosts.

The basic idea is to i3 service (overlay) mode is that sources send
packets to a logic identifier and receivers express interest in the
packets sent to the identifier. Receiver inserts a trigger (id, addr) to
express interest all packets in an identifier id. The packets sent to id
will be sent to addr specified. The significant challenge in implementing
i3 is to efficiently match identifiers. Simple idea is to match k-bits of
identifier instead of matching the exact identifier. This can be further
generalized to identifier stacks instead of a single identifier. More
importantly, i3 nodes manage the routing of these packets and end hosts
just need to know the nearby overlay nodes in the i3 network. Clearly, the
approach can be easily applied to mobility, multicast and anycast.

I3 is an overlay network in which every node (server) stores a subset of
triggers. If n triggers are generated all end hosts and if there are N
overlay nodes, each system on an average will store n/N subset of
triggers. I3 overlay system is implemented using chord lookup protocol. It
uses circular identifier space (0, 2^m-1).

Simulations were presented in the paper to evaluate efficiency of overlay
approach. The metric used for evaluation is latency stretch: ratio of
inter-node latency over overlay approach and ratio inter-node latency with
IP network. The results indicate that this approach is robust and scalable
and has a performance comparable to that of underlying IP network.

Key Conclusions:
1) Using chord lookup protocol, i3 system takes O (logN) hops to
route a packet to a server that stores the best matching trigger for the
packet irrespective of the starting point of the packet.
2) Public and private triggers provide security in i3 system.
3) Overlay approach is scalable, robust and provides scope for
incremental deployment.

Limitations:
1) I3 system requires that all id.s that agree in the first k-bits of the
identifier to be stored on the same i3 server. It is not clear how the
routing works in the case when id.s are split between multiple nodes.
2) Routing efficient in overlay network is a hard problem. It would be
interesting to see network dynamics of the overlay system with N nodes.
Received on Thu Nov 23 2006 - 11:00:14 EST