Internet Indirect Infrastructure

Paper Review: Internet Indirect Infrastructure

The paper proposes an overlay-based Internet Indirection
Infrastructure (i3) to act as a communication abstraction supporting
services such as multicast, anycast and mobility. The purpose of
their work is to generalize the point-to-point communication scheme of
the Internet.

In their model, indirection is provided when senders transmit packets
to a logical identifier and receivers express interest in packets sent
to an identifier. In contrast to IP multicast where the system is
responsible for routing, in i3 the end-hosts control the routing
providing flexibility, scalability and robustness.

Nodes can achieve mobility simply by updating the receiver address in
each of its existing triggers. Senders are not aware of mobility
since they continue to send to a logical identifier. Furthermore,
this abstraction enables mobility on both ends of the communication, a
major benefit of their scheme. Multicast is provided by having each
interested node register a trigger associated with their address. In
this manner, switching from unicast to multicast and vice versa is
straightforward through the registering and unregistering of triggers.
  Anycast is achieved by using inexact matching for the logical
identifier. A node can send a packet to an identifier matching in the
first k bits and this packet will then be forwarded to the receiver
with the longest matching prefix. By using identifier stacks instead
of a single identifier, a further abstraction allows for service
composition.

The service composition feature enables third party data processing
such as transcoding HTML to WML for mobile devices. Server selection
coupled with load balancing or locality can be provided by the anycast
feature since m-k bits can be used to encode preference or randomness.
  Large scale multicast can be accomplished by building a hierarchy of
triggers and forming tree structures. Their basic scheme clearly
provides enough functionality to support a wide variety of
applications which could not otherwise be realized with basic IP.

The use of public and private triggers enables both security as well
as session semantics. Their system can be implemented as an i3 proxy
on the end-hosts intercepting packets and translating them to i3.
Like IP, their model is a best-effort service. The paper introduces
an array of attack vulnerabilities but counters with proposed
solutions for each. Schemes such as challenges, loop detection and
fair queueing prevent malicious nodes from compromising the system.

The routing efficiency of the system was studied through simulations.
End-to-end latency was shown to converge quickly through the use of
sampling. Also, some simple heuristics are included and simulation
results show they provide performance improvements in the system.

This paper is impressive in two ways. First, it is written well and
properly explains the intricacies of the system without providing
excess implementation detail. Second, the project itself is
impressive from a design standpoint. In theory, the system seems to
cover all angles and does it both efficiently and completely. One
thing that is unclear from this paper is the robustness of the system.
  The authors claim that the system inherits the robustness of the
overlay being used. From my previous experience with Chord, it seems
that while it provides a fast lookup mechanism, it is not one of the
more robust topologies especially when scaling to large numbers of
nodes. Furthermore, the use of simple simulations and a barebones
implementation of the Chord protocol with only 32 nodes could have
been significantly improved. A full implementation is not out of the
question and could solidify their claims. Their results, however,
only serve as a proof of feasibility rather than a proof of efficiency
as stated in the paper.

-- Nadeem Abji
Received on Tue Nov 21 2006 - 11:46:43 EST