how mobile do we need to be (was Re: how mobile do we want to be)

[avri@psg.com]

On 22 mar 2005, at 17.22, Margaret Wasserman wrote:

> At 11:05 PM +0100 3/22/05, Iljitsch van Beijnum wrote:
> > That would require some serious voodoo as these are nearly always 
> > managed by different entities. So essentially this would keep the 
> > problem the same but require wifi base stations and gsm/umts 
> > infrastructure to solve this and hide the complexity from the hosts. 
> > This isn't traditionally the way the IETF does things.
>
> I was actually expecting that, to get this to work well, you'd need to 
> purchase the 3GPP/802.11 roaming service from a single provider...

I don't expect this to be the case in all cases.

In my view of things there is definitely a case where L3 could 
reasonably be involved.

What I expect to happen is that a 3G/802 equipped device/node would 
move into a zone where 802 was possible and  where at some point it 
could be connected to two or more providers at the same time, sort of 
multihomed. During this interval it could have separate addresses given 
by different providers. And as in any multihomed node it could chose 
which provider to use based on its own policy constraints.  As it moved 
from one zone's prevalence to another or from one provider's area to 
another, the balance would change and different providers, and L3 
addressing possibilities, could become available.

And while it is true that one could use the extensive machinery of MIP 
to make this work, it seems to me that using the multihoming shim's 
capabilities could be a lighter weight possibility.  But while it is 
possible that support for this sort of movement could just fall out of 
the shims specialized for static and fixed multihoming, I expect that 
support of a dynamic multihoming model, i.e. one with movement, might 
hold some gotchas.  I would prefer to see us deal with these during the 
architectural and protocol design stage then to have to deal with them 
in the deployment stage when the solutions might end up as 
afterthoughts, i.e addon hacks.

One thing I am pretty sure is that if it is possible to coerce shims 
into doing dynamic multihoming, i.e. multihoming with movement, then 
someone will.  Given that it probably will happen, it might be better 
to design for it.

a.