From: T Scott Saponas (ssaponas@cs.washington.edu)
Date: Mon Nov 15 2004 - 07:36:27 PST
Review by T. Scott Saponas
This paper proposes another solution to multicast, anycast, and mobile
networking over IP. This new comprehensive solution is called the
Internet Indirection Infrastructure or i3. i3 is an overlay network
approach that leverages existing IP switching below it in the protocol
stack. The idea is all of the above type of networking is really about
trying to create an indirection between sending and receiving. i3
creates this indirection by having senders send to an “id” over the
overlay network and clients receive by setting up triggers in the
overlay network for data with that “id” to be sent to them.
Some of the benefits to this approach are that it’s self-organizing and
can be incrementally deployed. The self-organizing comes from a
statistical multiplexing of the “id” used to identify data. With a long
enough “id” there is no need for a central registration because servers
can just pick a random “id” with a low likelihood of collision. i3 is
incrementally deployable (at least for non-mobile applications) because
only a few i3 servers must exist to create the initial overlay network
and only those servers providing content (like streaming news) and
interested clients would have to implement i3. One could imagine those
i3 servers being paid for by the same companies who are providing large
multicast content.
However, i3 has some drawbacks. For i3 to work in general for mobile
applications either every sever will have to start implementing it or
every mobile client will have to use a proxy. Also, it is not clear
that TCP applications will necessarily work well over i3. They show
there is some scalability to routing in the overlay and that pushing
triggers to other servers can keep any one node from being overloaded;
but it’s not clear from their simulations to what extent this can scale.
It was not tested with Internet sized networks and latency. Also,
while the authors do address some security concerns there is an overall
privacy concern. The authors assert that in most cases its no worse
than IP because to find out anything really interesting an i3 server
would have to be compromised. But I would argue since i3 servers seem
like they could be maintained by anyone and it’s the nature of the
system to be able to send any one flow of information multiple places
that there is the potential for many malicious people to fund/create i3
nodes for the sole purpose of monitoring traffic and selling that
information.
Despite the drawback addressed above I think the idea of a
scalable-self-organizing overlay network, like i3, as a comprehensive
solution to multicast, anycast and mobile computing over IP is a good
idea and i3 in particular shows promise.
This archive was generated by hypermail 2.1.6 : Mon Nov 15 2004 - 07:36:32 PST