From: Danny Wyatt (danny@cs.washington.edu)
Date: Mon Nov 15 2004 - 01:25:37 PST
Internet Indirection Infrastructure
Stoica, Adkins, Zhuang, Shenker, Surana
This paper proposes a new layer of abstraction, i3, that would sit
somewhere between IP and the application. i3 provides a decoupling of
sender from receiver by introducing a mediation similar to that proposed
for IP multicast: receivers register their address as a logical ID and
senders send to that logical ID, not the actual address. When a
receiver changes the address behind its logical ID, it has achieved
mobility. When many receivers register their addresses with the same
logical ID, they have subscribed to a multicast group. And by mapping
IDs in a prefix-hierarchy, multiple receivers behind a single ID prefix
can become an anycast group. The authors further extend the protocol to
add stacks of ID that allow for the chaining together of many receivers
that in turn allows for service composition (which could be anything
from media transcoding to virus scanning).
Generally, this protocol seems to collect and assemble similar
functionalities that have independently developed for many different
applications---from DHCP to DNS to SIP---in a shared abstraction below
the application. That said, it opens itself up to the same criticisms
applicable to anything that pushes against an end-to-end design: Does
is it cover all foreseeable application needs? Is it a worthwhile
replacement for most applications or just a lowest common denominator?
The authors do a very good job of considering many aspects of the
protocol (e.g., scalability, security, deployment), but the more
problems they consider the more ad hoc some of their solutions sound.
Additionally, large problems like triangle routing remain mostly open.
Ultimately, this seems to be a good unification of disparate efforts
into a common problem, but more refinements are probably worthwhile.
This archive was generated by hypermail 2.1.6 : Mon Nov 15 2004 - 01:25:43 PST