From: Charles Reis (creis@cs.washington.edu)
Date: Mon Nov 15 2004 - 01:34:21 PST
Internet Indirection Infrastructure
Stoica, Adkins, Zhuang, Shenker, Surana, 2002.
The paper attempts to address the problem of deploying new Internet communication mechanisms, such as multicast and mobility. The point-to-point nature of the current Internet keeps the network simple, but it also creates a great inertia that prevents new technologies from taking off. As a result, many workarounds use end-host designs (like end-system multicast) to avoid changing the core of the network. The Internet Indirection Infrastructure (i3) is designed to abstract out the common parts of these workarounds, creating a framework that adds a level of indirection between senders and receivers to support new technologies.
i3 provides an overlay network, which is recognized as an effective way to deploy new routing techniques on top of existing IP routing in an incrementally deployable manner. Moreover, it takes advantage of the strengths of structured P2P systems, such as high scalability, robustness to node failure, and self-organizing behavior. With a relatively simple technique of having senders and receivers register interest in each other via triggers with id stacks, very flexible designs can be deployed (eg. video streams with different formats for different receivers).
Unfortunately, many concerns (such as locality and load balance) start to complicate the design, and the system presents much easier opportunities for attackers to exploit the network, while the paper only begins to discuss options for handling such attacks. Perhaps most significantly, the authors are yet uncertain of the ability of i3 to be widely deployed, partly defeating the original intention. Nonetheless, small incremental deployments would still be useful to test new technologies without reinventing the wheel, hopefully with the ability to scale if successful. Overall, the simplicity of the approach (only providing the ability to send messages to those interested) is promising, rather than providing a complex API for all possible needs.
This archive was generated by hypermail 2.1.6 : Mon Nov 15 2004 - 01:34:26 PST