Re: [MEXT] About the flow binding charter item

[Julien Laganier <julien.laganier.ietf at googlemail.com>]

Hello,

<with my co-chair hat firmly off>

On Thu, May 28, 2009 at 1:25 AM, Giaretta, Gerardo
<gerardog at qualcomm.com> wrote:
>
> [...]
>> So, we would like that the WG express their opinion on both items
>> mentioned above:
>>
>> 1- whether we should replace Flow/binding policies exchange by Flow
>> Binding rules exchange in the charter
>>
>
> Agree

+1

>> 2- whether the information should be sent from the MN to the HA only
>> or it should also be sent from the HA to the MN and if the
>> information exchanged in each direction should affect the CoA and/or
>> the HA address.
>>
>
> From a protocol point of view it would be easy to allow both but from an architectural point of view we need to understand if that makes sense. As already commented there is no other entity in the picture which knows about the current application running in the terminal, the accesses available to the terminal and their status and characteristics, the resources available to the terminal than the terminal itself. The HA cannot know any of those elements which are needed to make a consistent flow routing decision. For this reason I think it should be unidirectional from the MN to HA.
>
> What the network can provide are pseudo-static information about how to use different accesses when they are available based on the environment. This kind of information is usually provided as configuration information and should not be carried over Mobile IP (e.g. OMA-DM is a more suitable protocol to configure such kind of policies in the terminal)
>

I agree with everything Gerardo said above, so no point in repeating
it. I'd like to add one point which in my view is key to this
discussion:

Functionally, the goal of the flow binding mechanism is to bind a
specific flow onto an access whose characteristics  (availability,
bandwidth, RTT, packet loss, etc. etc.) matches best the requirements
of this flow under the constraint of the current MN situation (e.g.
characteristics of the other accesses)

However here in the MEXT WG we are specifying a flow binding mechanism
for Mobile IPv6, and thus the flow binding rules are specified in
terms of CoA choice for a given flow -- they are NOT specified in
terms of which access to choose.

Now on the MN side choosing a CoA equals choosing an access, because
the MN knows onto which access a given CoA is configured. Thus, having
the MN establishing flow binding rules satisfies the goal of the flow
binding mechanism.

But if you look on the HA side, the HA has no idea onto which access
is a given CoA configured. Thus, having the HA establishing flow
binding rules DOES NOT satisfies the goal of the flow binding
mechanism, and is thus essentially useless.

Thus if someone is interested in having the network being in control
of the flow binding mechanism, that can't be done in terms of ruling a
specific CoA choice for a given flow, and thus that is unrelated to
MIPv6, and by derivation to MEXT. I don't know which place would be
more appropriate to tackle that issue, Gerardo mentionned OMA-DM, but
I can also see MIF addressing that sort of issue once they have
concluded the problem space analysis.

Thanks.

--julien