Re: [DNA] Issue about multiply connected to the same link and DNAbeing per-interface or not (was Re: WG LC comments ondraft-ietf-dna-goals-01.txt)

[JinHyeock Choi <jinchoe@gmail.com>]

Pekka

> >> For example, if a host has earlier been connected twice to the
> >> same link through two different interfaces, and then it detects
> >> that one of the interfaces has undergone a link change, IMHO that
> >> is a strong indication that there may have been a link change
> >> at the other interface, too.
> >
> > do you have a concrete example where this makes sense?
> 
> The specific example, though simplistic, that I've had in my mind
> is my home setting with my laptop.  I have a bridged Ethernet and
> WLAN.  When my laptop notices that I've disconnected it from the
> Ethernet, I may have just moving it around in our house, and it
> is still connected to the same link through WLAN.  On the other
> hand, I may have suspended it and moved it e.g. to the office,
> in which case it is now attached to a different WLAN.
> 
> Another example that I could imagine is a large operator network
> that from an IPv6 point of view acts as a single, large "LAN"
> or NBMA link.  (I know one operator that has plans like that.)
> The operator might offer different radio access technologies to
> access that "large LAN".  When one radio interface undergoes a
> change, it may make sense to quickly verify whether the other
> interfaces still are connected to the "large LAN".
> 
> In general, if two interfaces are connected to the same subnet,
> and one of the undergoes a link change, it often makes sense to
> check if the other one has also went through a link change.

I think the relation between a link change in one interface and 
a link change in another interface is similar to the relation between 
a link change in IP layer and a link-layer change in MAC layer. 
When a host changes its link-layer connection, it make sense to 
check whether a link change also has happened. However, 
a link-layer change doesn't automatically imply a link change, 
so IP layer can't solely depend on it and (typically) checks for 
link change for itself. So I guess their relationship is neither 
completely dependent nor completely independent, but, in between, 
interdependent. (It's as if I talk about the man-woman relationship. :-))

> > I don't have a problem relaxing the wording in the goals draft
> > to not accidentically preclude optimizations which share
> > information between interfaces, but I don't understand your
> > example above.
> 
> I was mostly objecting the notion that we should keep DNA on
> separate interfaces _completely_ independent.  To me, it looks
> like that there are cases when they are not.  I am oblivious
> how to handle this, through link hints, relaxing the language
> in the goals draft, or whatever.

It's not my intention to argue that we should keep DNA on 
separate interface _completely_ independent. 

It seems that we are in agreement that 

1. DNA schemes are typically run per interface.
2. When a host has multiple interfaces, the DNA related 
   informations can be shared between interfaces for the 
   better performance. DNA on separate interfaces need 
   not be completely independent. 

I think we have reached a common ground and would like to 
adopt your rewriting to resolve this issue. 

   DNA schemes are typically run per interface.  When a host has
   multiple interfaces, the host separately checks for link changes
   on each interface.

Kindly give your opinion on it. 

Best Regards

JinHyeock 

P.S.
This is 15th August, the greatest Korean holiday, HanGaWi. From 
earlier days, we celebrated good harvest with families under full
moon. Happy HanGaWi to all. :-)