Review: Internet Indirection Infrastructure

The Internet Indirection Infrastructure (I3), discussed in this paper,
is a means for allowing nodes to send messages to each other using a
logical address instead of a physical one. This scheme generalizes the
existing point-to-point communication paradigm that currently exists in
the Internet by decoupling the act of sending and the act of receiving
messages. Furthermore, this scheme provides an efficient framework on
which to build other communication services, such as multicast and
anycast.

One of the motivations for building a generalized indirection
infrastructure arises from the observation that specialized
communication services such as multicast and mobility contain common
elements in their solutions. However, the solutions themselves are
disjoint and do not lend themselves to be reused by other communication
services. The Internet Indirection Infrastructure provides a cleaner
and more reusable architecture for building these communication
services.

In order to send a message, senders use packets that are pairs (id,
data), where the 'id' is the logical address to which the packet is
being sent, and 'data' is the payload. Receivers, on the other hand,
use what are called 'triggers' to register their interest in receiving
a packet. A trigger consists of a pair (id, address), where the
receiver wants to receive any packets addressed to 'id', and 'address'
is the physical address of the receiver.

I3 servers form an overlay network, which is implemented using the
Chord lookup protocol. Whenever a sender wants to send a packet to some
destination 'id', the packet is routed to the I3 server responsible for
that id, and any triggers registered at that server are looked up. The
packet is then sent (via IP) to the physical destination.

The paper discusses some of aspects of I3 that any given implementation
must meet. These include, but are not limited to: robustness, efficient
routing, and avoiding hotspots. Robustness properties are supplied by
the overlay (Chord). Routing efficiency is helped by having the sender
cache the i3 server's IP address. Hotspots can occur when a large
number of clients try contacting a single I3 server, and is the result
of a trigger being overly popular; this can be avoided by caching, and
having an overloaded server push triggers to other servers.

The paper then goes on to perform some simulations and experiments with
a prototype of I3 that the authors built. The results are encouraging,
however the authors point out that their results should not be seen as
proof of efficiency, but only as proof as feasibility.

Overall, this was a good paper. It was well-written, and it provides an
interesting abstraction on top of which other communication services
can be developed.
Received on Wed Nov 22 2006 - 17:27:46 EST