Re: Survey on proposed IPv6 multi-homing mechanisms

[Mattias Pettersson <mattias.pettersson@era.ericsson.se>]

Francis Dupont wrote:
> 
>  In your previous mail you wrote:
> 
>    [cross-posting to mobile-ip list]
> 
> => please keep further messages in the multi6 list.

Ok.

> 
>    > Concerns
>    >
>    > - Needs Mobile IP implementation on destination host. So, modifications
>    >   in external hosts are needed to achieve internal multi-homing.
>    > - Mobile IP security mechanisms impose the use of authentication header,
>    >   raising complexity.
> 
>    Currently, Mobile IP WG has to come up with a solution to establish
>    security associations between the MN (host A) and the CN (host B). SA
>    establishment using public key infrastructure is assumed to not be
>    available everywhere and with all CNs.
> 
> => the mobility mechanism for IPv6 multi-homing was proposed before
> the security details were specified in the IPv6 mobility drafts...
> So don't expect that it works with any suggestion of this year (:-).

I know, just wanted to point out a realization problem.

> More seriously, the idea was that if the mechanism is applied to
> critical connections then these connections are already protected
> by some mechanisms, i.e. the key material is already available or
> very easy to setup.

Sounds highly visionary to me. Maybe it can be so.

> 
>    The (unofficially?) proposed draft-perkins-bake-00.txt suggests a (weak
>    but strong enough for the purpose) mechanism that involves a home agent
>    during the key exchange. The validity of the MN's binding of the home
>    address is verified by the CN by sending packets via the home network.
>    The home agent address belongs to the same prefix as the MN's home
>    address, i.e. PrefA:Prefsite::.
> 
>    If no SA is set up prior to the failure of ISPA, the key exchange will
>    fail since the CN can't send packets through the home agent any more.
> 
> => a SA set up prior to the failure of ISPA is the very easy case but
> can be common. Issues with the lack of home agent are different:
>  - the CN should not flush the binding (i.e. a bit/sub-option/...
>    is needed, its meaning should be "keep this binding as long as possible")
>  - there is no support for mobile to mobile, in the multi-home environment
>    this is translated by no support for failure on both the node and
>    its CN ISPs or one has to assume that this condition is detectable.
>    A common case is the failure of a high level ISP: if both nodes
>    are connected to it they will see a failure at the same time,
>    to understand this situation is hard but not impossible...

  - Global PKI...?

> I like this mechanism (of course :-) but we wait for the IPv6 mobility RFC
> before work again on it. BTW this is very easy to implement (movement
> detection is a dream).

I agree that mobility has powerful mechanisms to overcome "small"
problems such as multihoming... ;)

Still some work has to be done because this won't work out of the box.

/Mattias