Re: [DNA] DNA BCP Submission

[JinHyeock Choi <jinchoe@samsung.com>]

Dear Sathya

> >I agree that 'RA with current active prefix' can be used for link 
> >identity detection but kindly take notice an adverb 'PROPERLY'. 
> >That's why I wrote that it can't be a link identifier, NOT it can't 
> >do link identity detection. 
> >  
> >
> ok. And, my point is that as long as you can do link identity detection 
> even though you don't define a link
> identifier it is still a good way to go.

Agree. If only link identity detection can be done quickly and precisely, 
we don't have to adhere to link identifier approach. . 
 
> >If a host receives an RA with its current active prefix, it means it still 
> >remains at the same link. However, if a host receives 'an RA with 
> >a different prefix', it can't check for link change with it. 
> >  
> >
> A timeout without receiving a 'RA with current active prefix' can be 
> used to indicate link change. Right?

What concerns me about timeout approach is 
1) it's slow because a host needs to wait till timeout.
2) it's not precise. You need to rely on approximate knowledge, 
because other incidents, such as packet loss or 
prefix omission can occur timeout without receiving prefix.  

> >We expect 'link identifier' to EXPLICITLY represent a link identity
> >and have the following two properties. 
> >1) When the same link identifer arrives, it means the same link.
> >2) When a different link identifer arrives, it means a different link.
> >  
> >
> I agree with these two properties with one 'little' caveat - these two 
> properties need to hold only for a
> particular node in question. i.e. two nodes in the same link could be 
> using different 'identifiers' to represent
> the link as long as within each of them the above two properties hold. 
> You seem to be making
> his point below, but ...

Agree. But I am afraid that in a link, unless all the node that use 
the same technique agree on the same link identifier, these two 
properties are very difficult to hold even for a particular node. 
Assume node1 and node2 use the same link identifier technique, 
hence the same link identifier format. If node1 & node2 use 
different link identifiers, when node2 receives the link identifier
used by node1, even though it receives a different link identifier,
it doesn't mean a link change. 

This is an example w.r.t. RRD. In a link, assume node1 uses prefix A:: 
and node2 Prefix B::. When node1 receives an RA only with prefix B::., 
though it receives a different prefix, it doesn't change a link.  

> >To have the above properties, all the nodes adopting the same 
> >link identifier method, whether it be CPL or LinkID, should agree 
> >on the same, hence unique, link identifier. For example, if in a 
> >link, a node use CPL and the other use LinkID, then they may 
> >have different link identifers. However, among the nodes using CPL,
> >or any other link identifier, there can be only one link identifier
> >for a link. Otherwise 2) can't be satisfied.
> >  
> >
> even within a same scheme , two nodes using the scheme could use 
> different 'identifier', as long as the
> above two properties hold within the node (neither CPL nor LinkID is an 
> example for this). '

I wrote my opinion about this on above. Also I think, 
without the unique link identifier in a link, to satisfy those two, 
each link identifier needs to carry information that indicates for which 
nodes this link identifier is intended. 

> I guess,
> you know where I am going with this. In my proposed solution draft, RRD: 
> I am trying to use the current
> prefix and force the timeout within a short duration by defining new 
> timer for the current router. Even though
> it doesn't define a 'unique' identifier for every node in the link, I 
> think it is reasonable approach to achieve
> link identity detection, with its own advantages (and ofcourse, 
> dis-advantages).
 
I admit that my knowledge on RRD is not thorough. However, it's not 
clear to me 
1) how to quicken timeout without side effects, such as false link change detection. 
2) what's the advantage of using <current prefix, current access router>, 
     instead of just a global prefix. 

If you use timeout to check for link change, any global prefix in a link can do 
the same, w.r.t DNA. 

A host remembers any prefix in a link, when it suspects it may be in a different 
link, it sends an RS. If it receives the same prefix, it's in the same link. If it doesn't 
receives it until timeout, a link change has happened. It doesn't have worry about 
'current' things. 

Best Regards

JinHyeock