RE: Host Centric Multi-Homing

["marcelo bagnulo" <mbagnulo@ing.uc3m.es>]

Hi Cedric,

thank you very much for your feedback and your interest


>
> 6.4 Solution 4: Host based exit path selection + source address based
>     routing
>
>  "Required modifications: implementation of the path discovery
>   mechanism"
>
>  What is exactly the path discovery mechanism ? Is it the path failure
>  detection described 5.1.2.2.2 ?

Yes

> You should perhaps refer to the
>  appropriate section where the mechanism is described.
>

I will do that

> 7. Proposed solution
> 7.1 Multihoming solutions for small sites
>
> I don't see clearly which solution (from the ones you describe in
> section 6) you recommend.

Yes, this is not clear

the proposed solution is a combination of some of the solutions

> In fact, you are recommending 2 (optimized) solutions for small sites :
>
> - Host based exit path selection + site exit discovery
> - Host based exit path selection + source address based routing
>

This is correct.
the basic solution is the one based in source address routing and the site
exit discovery is an optimization

> Some mechanisms you discuss in section 7.1 apply to the first solution
> while others apply to the second. This is a little bit confusing, and it
> is difficult to have a coherent view of each solution you propose.
>
> You should perhaps present the mechanisms separately for each solution.
>

Yes, it is confusing, i agree

the point is that section 6 is an abstract analysis of the options, while
section 7 is an attempt of solution. The optimal solution is a combination
of multiple mecahnisms, as you have already pointed out.

But i will try to improve the readability of the doc.

> BTW, it would be interested to know which solution (based on source
> routing or based on site exit discovery) is preferred by you and by
> people out of this wg. There is an important design choice to make here.
>

IMHO the adoption of a multihomed solution can be proposed as a three step
process.

The different steps require different amount of modifications

1- the first step would be source address based routing restricted to the
site exit routers (a mesh of tunnels between the site exit routers). This si
simple, and required no upgrading (neither on hosts nor in routers)
This first step overcomes the site exit issue and allows regular (non
upgraded) IPv6 hosts to work propoerly in a multihomed site. The problem
here is that the routing of the packets will be suboptimal

2- the second step is to modify the hosts to make the site exit router
discovery. If so, upgraded hosts would obtain an optimized performance,
since they would directly send the packets to the appropriate exitr router.
This way we achieve optimal (internal) routing. The problem here is that we
still introduce overhead

3- The third step woulñd be based in source address based routing. This may
require a upgrade of the routers and perhaps routing protocol. At least it
requires to understand better what is needed to achieve this, and why people
don't like it


>
> 7.2. Multihoming solution for medium sites
>
> Which solution from section 6 are you recommending ??

In this case the hybrid approach is proposed. Path availability is
discovered using BGP and the site exit routers inform the host which is the
correct isp (source address) that they should use (something like putting a
NAROS server in the exit router)

> >From what you described, it seems that it is solution 5 (host
> based exit path selection + site exit discovery). But this is
> contradicting since you said solution 5 is only suitable for small
> sites.
>
> I also deduce you are not recommending source based routing. Maybe
> you should make this point more explicit.

Well, you could use source address routing or let the host tunnel the packet
directly to the host. The three step roadmap would also fit here.

>
>
> -------------------------------
> Purely editorial
>

Ok, thanks

Regards, marcelo