Re: [DNA] Re: Flash renumbering

[Erik Nordmark <erik.nordmark@sun.com>]

Sathya Narayanan wrote:

> I know this is always an issue. But, isn't it a general problem about
> CompleteRA which we need to think about either way?

In other cases we have the CPL logic ("wait for 4 seconds") to fall back 
on when the complete bit isn't set.

I suspect in you suggestion we can have the routers handle things 
without a complete bit; if a router hasn't heard a learned prefix from 
any other router, then after 1.5 hours it can drop it from the learned 
prefixes.

Thus the issue now becomes whether and how the router can communicate to 
the hosts that this is no longer a (learned) prefix on the link. We can 
overload the complete bit with this when the prefixes all fit in one RA, 
but we could perhaps also do some explicit "removal of learned prefix" 
type of option??

> I don't remember why we decided to remove the #prefixes field from the
> DNAO.  I thought most of us felt that this field will help out with
> being able to send the prefixes in two or more RA if they don't fit in a
> single RA.

Sure you can send them, but that doesn't mean we know what the semantics 
need to be on the receiver.

> So, let me bring #prefixes back instead of the complete bit, here is
> what I propose when the prefixes don't fit in one RA:
> 1) Divide the prefixes into groups with atleast one common prefix so
> that each group will fit in a single RA.
> 2) Set the #prefixes field in the DNAO to the total number of prefixes
> (The semantics for this is "This is the number of prefixes on the link
> *I* know of").

I don't think #2 can work in a robust manner unless we apply normal 
fragmentation/reassembly to such large RAs. And I don't think we need 
this complexity.

   Erik