Re: [DNA] Couple of points on draft-jinchoi-dna-soln-frame-00.txt

[Greg Daley <greg.daley@eng.monash.edu.au>]

Hi JinHyeock and Alper,

JinHyeock Choi wrote:
[cut]
> 
>>  2.  We assign each link a locally unique Link Identifier.  'Locally
>>      unique' means, no two adjacent links have the same Link
>>      Identifier.
>>
>>This would have issues if the hosts ever make a long jump (skipping some
>>intermediate links) for any reason.
> 
> 
> The term 'adjacent' needs further contemplation. That's one of why I wrote that
> there are complexities related to 'local uniqueness'. We may make it more clear
> when we discuss the idea of 'Link Identifer'.  

Indeed, Brett and I have discussed this with JinHyeock before.
the local vs global identity issue is a matter of appropriate tradeoffs.

(whether the LinkID is small enough to be included in all RAs,
whether the LinkID is unique enough, what constitutes adjacency).

In the end I think the idea that we agreed on was that any network
which you could transition from or to is adjacent (even long leaps).
This may push up the number of adjacent networks particularly in
the case of wireless networks.

If we're using non-globally-unique identifiers (rather attempting to
get local uniqueness), it is likely that we won't explicitly be
aware of all of the possible adjacent networks.

The fact that we're not likely to know the adjacent networks means
that we shouldn't really rely upon manual configuration (since everyone
starts link numbering at 1...).

Therefore, any link idenfifier which isn't globally unique needs to
adopt (as close to really random as possible) random identifiers.
Therefore, when analysing random identifiers, it may be necessary to
take a conservative approach to calculating probabilities of adjacent
linkID collisions, and potentially rectifying them.

Global uniqueness brings its own challenges

Uniqueness is likely to be guaranteed by the routing
infrastructure if we use a prefix, and prefixes may change.
Additionally, prefixes have their own semantics, which may
prevent PIOs being used as linkIDs.
Another option which contained all of a 64bit prefix wouldn't
fit into the smallest ND option size, and would require
16 octets (twice the size of a 48bit random linkid in an
8 octet option).

It's up to the group to determine if that's too big to
include in every RA, or if that's a goal

Greg