Re: [DNA-BOF] Using addresses optimistically and optimizing DAD

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

Hi Pekka,

Pekka Nikander wrote:
> Greg and Youn-Hee,
> 
> [Changed the subject to match the topic discussed.]
> 
> Greg Daley wrote:
> 
>> I'm not sure if we can use the global addres optimistically
>> or not (I haven't finished thinking about it).
> 
> 
> I wish we could, i.e., I see certain needs for such functionality.
> Consequently, I think that we should produce an analysis of the
> potentially adverse effects of sending packets, destined to
> a global address and therefore a stale router link layer address,
> and sourced from a global address.  My current understanding is
> that in most cases the link layer drops those packets, and no
> real harm is done.  In the improbable but possible case where there
> happens to be a node whose link layer address is the same
> as the router link layer address from the previous link, that
> node will get some traffic not destined to it, and the IP layer
> will drop it.


There are links where link-layer addresses are guaranteed to only
be unique on a per link/cell basis.

In this case, the router on a new link-instance may receive
the packet (for example with TEI=0), but existing
specifications indicate that they should ingress filter the
packet instead of forwarding it.

> Hence, to me it looks like the only result of using such
> addresses is some unnecessary traffic.  If we rate limit
> the use of such addresses, I don't see much problem.
> But I probably fail to see something.

Upon reflection, I think you're correct.

If we've got active connections which are continuing to send data,
with global source addresses, there will be no chance of DAD
collision if we move to another network which is not using this
prefix.  If we're on the same subnet as before, we already know
that the address is ours.

Modifications to ND state aren't needed here (in fact, we
may prevent outbound packets causing NS until DNA is complete,
not sure).

In this case, the packets to be going onto the link are
DNA packets, DAD optimization packets for link-local and
existing sessions with non-link local traffic.

Per host or interface configuration capabilities could
disable this behaviour if necessary.

Once link-local DAD completes, or allows packet transfer,
further operations may be allowed (including ND, although
this would benefit from DNA completion).

>> I'm pretty sure that we shouldn't do DAD for a global
>> address unless the link had changed (and the global address
>> wasn't valid). This implies DNA completion (although
>> configuring a new address may be applicable as soon as a new
>> prefix is received).
> 
> 
> I think I agree, but I am not sure (still thinking).
> 
>> For the link-local address, I think we can say that we don't
>> need to wait to hear what DNA says, since I may want to use an
>> Optimized DAD procedure so I have a link-local address for use
>> in DNA (for neighbour discovery for example).
> 
> 
> Yes.
> 
> --Pekka Nikander
> 
> 

I think we've got similar pictures in mind, now.

Greg