From: Jenny Liu (jen@cs.washington.edu)
Date: Sun Nov 14 2004 - 18:08:03 PST
The authors of "Internet Indirection Infrastructure" present a new
communication abstraction that allows different classes of communication
services to be easily built on top of it.
The proposed abstraction infrastructure is great. It's scalable,
efficient, and self-organizing. In particular, it solves (all at once)
many of the problems associated with implementing multicast, anycast,
and mobility over the traditional unicast over IP infrastructure.
However, i3 makes no guarantees about the length/cost of the path from
the sender to the receiver. In fact, it puts much of the burden for
ensuring that a short/low-cost path is chosen onto the receiver (who may
not know how to do so, and) which makes it possible for a malicious
receiver to purposefully try to obtain long/high-cost paths from the
sender, thus unnecessarily contributing to congestion.
The paper says that existing UDP-based applications can work over i3
easily, but it's not clear that TCP-based applications would do well.
However, it seems that most multicast applications (real-time
video/audio, data distribution) would do fine over UDP, and the authors
present mobility support for TCP-based applications in a separate
paper. However, there is no talk of TCP support for anycast, which
might be necessary depending on the anycast application.
Another question that remains is "who would pay for these i3 servers?"
Though the infrastructure may make further development of communication
serivces easier, maybe the services that have already been developed on
top of the traditional infrastructure are good enough.
i3 sounds great. If people are willing to invest the cost of switching
over to i3, it could mollify the current situation of disjointed
communication services.
This archive was generated by hypermail 2.1.6 : Sun Nov 14 2004 - 18:07:55 PST