Summary: TCP Migration

The paper presents a end to end architecture for Internet host mobility
that makes no changes to underlying IP substrate. The paper makes the case
using the end to end argument that mobility might be best implemented as a
higher level, end to end function like reliability.

Inspired by akamai, the authors take the advantage of DNS and it's ability
to support secure dynamic updates and its ability to support secure
dynamic updates to locate the mobile host's network attachment point. The
end to end host mobility presented here requires a DNS entry and a shared
connection between two end hosts. It scheme has three important
components:
1)Addressing: Mobile host users locally obtained address in the foreign
network.
2)Locating Mobile host: If mobile host is acting as a client, no updates
needs to sent to DNS. Whenever mobile host changes its IP, it must detect
this change and change the host name-to-address mapping in DNS.
3)TCP Migration: This is achieved by is adding an TCP option to support
migration. Secure migration is ache vied by using DHKE during the initial
connection establishment. Whenever mobile host changes, the fixed host
goes to MIGRATE_WAIT state till the mobiles comes back to the network in
another foreign network.

Since the scheme presented in the paper does not impose any triangular
routing anamolies, the end to end latency for the scheme presented here is
better than standard mobile IP and similar to Mobile IP with route
optimization. The paper also addresses some security issues related to Dos
attacks and connection hijacking.

Overall, I thought idea presented in the paper is very good. I felt the
paper was accepted for the conference because of idea but not because of
their implementation/results. I felt authors should have spent more time
on doing the evaluation of their architecture. Especially, I would like to
see in practice how big is the problem of two hosts moving
simultaneously.

As suggested in the paper, the architecture has few limitations. Firstly,
the architecture does not support if both the hosts are moving
simultaneously. This happens when intervals between address change and the
reception of migrate SYN by the corresponding host for both end hosts
overlap. This limitation basically stems from the fact that there is no
anchor point like Mobile IP's home agent and any IP address change has to
happen before the other host can proceed. Secondly, it requires changes to
every end host i.e. each transport protocol. The paper also does not
present how the architecture can be generalized for UDP-based protocols,
though it states that it can be done with little difficulty.
Received on Thu Oct 19 2006 - 11:03:23 EDT