From: Tom Christiansen (tomchr@ee.washington.edu)
Date: Wed Nov 17 2004 - 16:26:40 PST
This paper presents an overlay or addition to IP. Originally, IP was
invented to allow p2p communication between two nodes on an internetwork.
However, IP multicast remains a bit clunky - especially for mobile users
who change IP addresses as they move and, thus, log on to different
networks. The presented system, therefore, abstracts from the IP addresses
and invents an identity which is broadcast onto the network in form of a
trigger. In its simplest form, this trigger is the node identifier and its
IP address, port number. As a result a change in IP for the receiver is
completely transparent to the sender as it only deals with the trigger ID.
The protocol has a few strong points going for it:
- It increases flexibility compared to IP multicast.
- It is incrementally deployable.
- It is virtually unlimited in terms of types of services, data, etc. the
protocol can be used for.
- It is backwards compatible with existing protocols (running UDP datagrams
through i3 is mentioned as an example).
Unfortunately, i3 also has some drawbacks. The most serious ones are the
security issues described in the article. But these issues exist with
TCP/IP as well.
The article briefly mentions that hot-spots are to be avoided with this
protocol. CNN is mentioned as an example. A server with many users will
either have to 'advertise' itself using multiple triggers (which would
break the IP abstraction) or just have to deliver new triggers at an
alarming rate. This is probably the weakest spot in the protocol. It would
have been nice with a more in-depth description of it.
Overall, a well-written article. Relatively easy to read. Good, informative
figures.
This archive was generated by hypermail 2.1.6 : Wed Nov 17 2004 - 16:27:09 PST