Re: [DNA-BOF] BoF Last Call on DNA WG Charter.

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

Hi JinHyeock, Bernard and Alper,

We've got a few stock replies so far indicating that
Optimistic DAD may help this issue.

Can anyone see a problem with attempting to do
Optimistic DAD on a previously configured address
either:

1) for link-local (v6) configuration?
2) after receiving a hint (or confirmation,
    depending on the strength of the hint)
    that we've returned to a network where we have
    an address configured?

Personally I guess that the main issues are:

* That we've got a situation where hosts can have
   short disconnections every few seconds, (so we may
   have too much traffic)

* That optimistic DAD has insufficent safety in the
   case of a collision (<hatoff> though I've read the
   draft and don't think this the case</hatoff>)

* That it's not worth the effort to specify this until
   RFC2462bis is more concrete.

Any opinions or additions?

Greg

JinHyeock Choi wrote:
> Hi Alper
> 
> Alper Yegin wrote: 
> 
>>>The same problem occurs with IPv4 Link-Local addresses.  Since the
>>>probability of conflict is relatively high, we had to recommend that a
>>>host awakening from sleep, attaching, etc. had to do DADv4 again.
>>>
>>>The chance of conflict is much lower in the case of IPv6, so I'd ask
>>>whether we think that the same approach is still necessary.  It's quite
>>>costly, of course, so if it can be avoided that is a good thing.
>>
>>This is a hard question. The probability of collision (p1) is somewhat not
>>low enough, hence the RFC 2462 requires use of DAD. Now we are talking about
>>the possibility of an address collision sometime after the address was
>>checked out to be unique (p2). This is even a lower probabibility, yet how
>>do we determine it is low enough to be negligible? Also, note that the p2 is
>>a function of the number of new IP addresses configured while the host was
>>not defending its IP address. This is higher in a network where the number
>>of hosts joining the network (i.e., mobility) is high.
> 
> 
> It also is a function of the time during which the host is absent. It will be 
> difficult to set a criterion for 'to DAD or not to DAD'. :-) 
> 
> The possible solution may be not to set the criterion but to simply use 
> 'Optimistic DAD' after reattachment. 
> 
> Best regards
> 
> JinHyeock  
>