An End-to-End Approach to Host Mobility

Paper Review: An End-to-End Approach to Host Mobility

The paper presents a novel approach to enable host mobility on the
Internet, specifically the migration of existing TCP connections
without a need for a third party. An end-to-end approach was taken
rather than a network-layer scheme as is the case with mobile IP. The
paper argues that this method provides ease of deployment and better
performance. To locate hosts which have moved, the system uses the
Domain Name System (DNS) to update the name-to-address mapping in the
host?s home domain.

In their system, a host obtains an IP address valid in the foreign
domain it is migrating to. This has several benefits in that it
doesn?t violate any trust issues, maintains the scalability of
Internet routing due to the IP addressing scheme and does not require
any triangle-routing or reverse tunnelling. Unless a new system is
truly groundbreaking, the fewer the modifications it makes to the
TCP/IP model, the greater the likelihood it has to being implemented.
Their scheme exploits the abstraction provided by DNS. A host name is
not required to have any geographical connotation and thus can be used
as an identifier even when the host is mobile.

The paper brings up a possible race-condition for the case where a
mobile host migrates after a host trying to contact it has retrieved a
now invalid address from DNS. The solution stated is to have
application-level support. The authors seem to skirt around this
issue as this goes against the goal of making the mobility
transparent. They are lucky in the sense that these race conditions
are a rare occurrence and the infrequent errors they cause have
acceptable consequences.

To allow for migration of TCP connections a new Migrate TCP option is
included in SYN segments. This is necessary since each TCP connection
is associated with a source and destination address port 4-tuple which
naturally does not allow migration. Hijacked connections are guarded
against by utilizing a secure token. Even though IPv6?s IPSec
provides the means for securing a token, the author?s present their
own scheme since it is unclear when and if IPv6 will become the
standard. A MIGRATE_WAIT state is introduced to handle the situation
where mobile hosts are in the process of migrating and have not yet
reconnected from a new location. The paper suggests a passive
approach to migration, which is a good idea, since mobile hosts may be
on the go before they have a chance to notify the receiver. This
methodology is clearly more generic and will be able to support more
applications varying in nature. Mobile hosts are able to signal an
impending migration, but are not required to.

One noted advantage of handling mobility end-to-end is that it allows
the higher layers to be aware of the fact that the host has relocated.
  This could, for example, allow TCP to begin in slow-start after a
location change to adapt to new network conditions. Their scheme also
does not require any changes to TCP headers, packet format or semantics.

The authors seem to go back and forth on certain issues such as
performance enhancements, security measures, deployment and address
caching leaving the impression that they are not fully clear
themselves on some design aspects. Also, for a paper suggesting a
modification to the TCP protocol, they could have conducted some
extensive experiments rather than their very basic one. This would
lend more credibility to statements like their scheme provides better
performance than mobile IP. Aside from these issues the paper has
some very useful ideas. First, an end-to-end approach to host
mobility is both feasible and perhaps even the more natural
methodology. Second, there are several issues which must be taken
into consideration especially involving security when designing such a
system. Third, it is important to leave the system as generic as
possible to support applications which vary in nature which is similar
to one of the goals of the Internet. Fourth, the DNS system can be
used to provide a simple form of identity in the Internet required for
mobility. Finally, the implications of IPv6 must be considered in the
design of new systems.

-- Nadeem Abji
Received on Wed Oct 18 2006 - 02:52:36 EDT