Re: Transport multihoming

[Manuel Urueña Pascual <muruenya@it.uc3m.es>]

> Manuel Urueña Pascual wrote:
> > Hi,
> > 
> > I've been working in a protocol that seems very close to your reasoning.
> > Not really a new approach but catching some good ideas from SCTP and
> > Mobile-IP to exchange dynamic/multiple addresses between end-hosts with
> > legacy transport protocols.
> > 
> > http://www.it.uc3m.es/muruenya/draft-muruenya-epcp-00.txt
> > 
> > I think it may be applicable to the multihoming problem, but I'm seeking
> > some feedback from the MH people.
> 
> Very interesting, IMHO.  Architecturally it looks fairly similar to HIP.
> What would you consider to be the main differences?

I agree with you, they are very similar, and IMHO any end-to-end MH
solution will look the same.

I've only read a paper by Francis Dupont so my knowledge about HIP is
far from deep, but I'll try:

EPCP does not require any kind of crypto capabilities or infrastructure
for basic association init, maybe for cookie generation, as it relies on
the Return Routability check. However, it allows any other mechanism to
authenticate an end-point address. In fact, CGAs are suggested as an
alternative mechanism in the Fast Add section.

I'm not really sure about this (please correct me), but HIP seems to
support just one address, that may be changed for another one with the
Readdress packet. Maybe that's a hard constrain, EPCP supports several
simultaneous addresses, even IPv4 and IPv6 altogether.

Maybe the main difference is my perception about the Identifier -
Locator problem. I think IP addresses are great identifiers: global
unique, fixed length... (see more properties in "Endpoints and Endpoint
Names", J. Noel Chiappa), and much more important, today's applications
employ them. That's why Primary Address is employed as an identifier
mapped to several locators (even to itself).

--Manuel