Internet Indirection Infrastructure

I. Stoica, D. Adkins, S. Zhuang, S. Shenker, S. Surana, "Internet Indirection Infrastructure," ACM SIGCOMM Conference, (August 2002).

 This paper attempts to generalize the Internet's point to point communication system. Traditionally, Internet has always been based on a push model, wherein the sender pushes the packet to the given receiver by explicitly addressing the packet to it. i3 on the other hand, proposes what we can think of as a push-pull model where the sender pushes a packet in the network and it is the responsibility of the receiver to pull it towards it.

Broadly, the idea is that each packet is associated with an identifier which is used by the receiver to obtain delivery of the packet. The sender sends the packets as (id, data) where id is an m-bit identifier and data is the payload. This packet is directed to one of the many i3 servers. The receiver on the other hand, use triggers of the form (id, addr) to indicate their interest in a packet. It is the job of the i3 servers which form a part of the overlay network to match the identifiers and forward the packet to the given addr. The authors also proposed a second generalization of i3 which consisted of sender sending (id_stack, data) and the trigger comprising of (id, id_stack) and the intermediate i3 servers implemented in a chord structure wherein the packets get hashed according to the prefix function of the first k bits of their m-bit identifiers.

Based on this, the authors showed that a variety of services can be implemented efficiently. Primarily applications that involve mobile receivers can now be implemented with ease as the receivers can update the i3 servers on the fly whenever they move to new locations by creating new triggers. Further multicast and anycast can be implemented similarly without the sender being concerned about the individual addresses of receivers. The authors also showed that strong applications such as service composition can be implemented with the help of indirection servers in the middle without the knowledge of the sender, which I feel was the main strength of the paper.

Comments

Overall, the paper builds on a definitely great idea! The whole push-pull model was nicely explained and make total intuitive sense. However, some issues that come to mind are:

1. I3 like all other overlay networks suffers from latency stretch. That is, every packet is routed to i3 servers instead being directly sent to the receivers. Though the authors have proposed a number of steps that could really localize the nearby i3 server where the packet is stored, it would always suffer from higher latencies as compared to sending directly to the receiver.
2. I felt that the challenge based protocol to prevent DDoS attacks which consisted of sending a nonce that is expected to be returned by the receiver was too simplistic and can easily be broken.
3. The authors donot really talk much about the time they should store the packet on each server, the trigger validity period etc. These values are crucial if i3 aims to be deployed on a large scale.