Re: Use cases for optimized DAD (was Re: [DNA-BOF] Do we need aseparate DAD Problem Statement?)

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

Hi Pekka,

Pekka Nikander wrote:
> Soohong Daniel Park wrote:
> 
>> Optimized DAD procedure;
>> 1) optimized DAD for link-local address before DNA
>> 2) obtaining link-local address and completing DNA
>> 3) obtaining global address informations (prefix...)
>> 4) composing global address
>> 5) optimized DAD for global address ( or skipping)
>>
>> Am I missing something ?
> 
> 
> I am thinking about a situation where a moving node can
> anticipate, with high probability, the prefix that
> it is likely to use on the new link.  For example, you could
> say that the prefixes used on a given wide are WLAN network
> were
> 
>   <common prefix>:hash(base station link layer address)::
> 
> If that was the case, the node could optimistically assume
> what its prefix is likely to be, and optimistically
> configure a global address even before it has sent any
> IP packets on the new link.

I know this is jumping off the topic, but...

I think that this design interesting, but may not be generally
applicable (we have only 16 bits of network prefix which could
be used for the hash).

It also constrains us to a single router/network per base station.

In these cases, eager cell switching based on RA reception
will work nearly as well (there are systems proposed which can
consistently get an RA within < 20 ms).

I understand that this idea was for illustrating the point
with DAD/DNA interaction, rather than as a complete
solution, though.

> This more or less falls withing your framework, but with
> the twist that 1) and 5) can happen at the same time, i.e.,
> before DNA has completed.
> 
> Summary:  I would rather not see any unnecessary design
> barriers for optimized DAD.  I merely see the two distinct
> cases of more or less random addresses (which can use an
> optimistic DAD procedure) and stateful addresses (which
> can use a procedure based on pre-allocation).

I think that if we have to consider how global addresses
can be used during DNA, then we're bound to come up with
these questions.

In most cases though, (as discussed before on the list) the
potential ill-effect of transmitting with the old global
address is limited to bandwidth utilization on the new link,
and some packet loss.

For designing DAD goals, I'd say that the scheme should to be
applicable for global and link-local addresses.  Also,
it should be able to be used whether or not DNA is
completed (although DAD on an off-topology global address
is likely to give misleading results though).

Ordering need not be strict, if it doesn't lead to
false behaviour.

Greg