From: Alan L. Liu (aliu@cs.washington.edu)
Date: Mon Nov 15 2004 - 01:00:39 PST
Instead of placing the burden of managing unicast, multicast, and
anycast flows on the end host, the paper proposes i3, an indirection
service that provides a simple interface for all three.
The paper presents many ways in which i3 manages what used to be managed
by end hosts or applications, such as longest prefix matching to support
anycast. It also identifies a key area where i3 is better than plain old
IP even for unicast -- mobile hosts.
Some other points it manages to bring up about i3 include security for
avoiding malicious redirection of triggers. The paper makes clear that
as an initial proposal there may be more things that can be done later
on, but they do give solutions that allow i3 to meet or exceed the
Internet standards for service and security.
I'm not clear from the evaluation how much of a performance penalty
using i3 is compared to standard IP for unicast, multicast, and anycast,
especially since they did not provide numbers for multicast. ALthough
they compare latency between i3 and Internet in the packet stretching
graphs, they do not compare throughput. They also dismiss analyzing
control message overhead simply because they used few servers. Finally,
their infrastructure security scheme merely prevent malicious triggers
from setting up a loop, but does nothing about a host that could set up
a long chain to essentially waste network bandwidth.
The main things the paper does a poor job explaining about i3 are
accountability and management. How do competing ISPs make sure i3
doesn't create traffic they don't want to pay for? What could a network
administrator do to make sure the part of i3 they support is working?
i3's additional level of indirection comes at the cost of less control,
which is both a blessing (allowing incremental scalability) and a curse.
In the end, i3 seems like an interesting layer on top of IP, but it
seems geared towards niche needs like anycast and mobile hosts. It's
unnecessary for the vast majority of current bandwidth since it doesn't
provide a significant benefit to fixed unicast communication. Perhaps in
the future i3 or something similar will replace the ad-hoc mish-mash
approach to dealing with differing routing needs.
This archive was generated by hypermail 2.1.6 : Mon Nov 15 2004 - 01:00:42 PST