Re: [DNA-BOF] Attachment Detection charter text for discussion

[Brett Pentland <brett.pentland@monash.edu>]

Bernard Aboba wrote:
>>Perhaps my wording is not so good after all.  I simply meant that
>>"network attachment" occurs when an interface becomes active on a
>>link instance.
> 
> 
> What does "active" mean?  When the interface senses carrier? Or when all
> tasks necessary to make the interface ready for L3 connectivity are
> completed (e.g. L2 authentication & association, etc.

Yeah, good point.  I meant the latter but that was not obvious.  I think 
that at that instant "network attachment" has occurred (others may have 
their own ideas), but I think that the job of detecting network 
attachment may go beyond that instant.  Even a "perfect" link-up trigger 
that doesn't provide any extra information just tells you that you've 
attached to something, but not what that network is.  I think that part 
of the job in detecting network attachment is finding out which network 
you have attached to and whether that means that further action is 
required (like configuring a new address).

>>How that is detected is the work of DNA.
> 
> 
> It may not be easy to "detect" at L3 when the interface has
> not yet entered an L2 state in which L3 connectivity is possible.  If the L2 state is
> unauthenticated, it can look to L3 like there is no connectivity at all,
> leading to all kinds of poor decisions, such as allocation of an IPv4
> Link-Local address.

Agreed.

> Without L2 "hints", the host will not receive an RA,
> and may attempt to test reachability to the default gateway or send an RS,
> all to no avail.  By the time that the L2 state is truly usable, the
> interface state may have been backed off considerably.
> 
I was thinking in terms of mobility, where with no L2 hints, L3 will be 
in the dark (thinking it's still connected to its old network) until an 
unsolicited RA arrives, some time after L3 connectivity has occurred. 
Are you referring to a host that is coming up for the first time or has 
been disconnected for an extended period?  I guess in that situation, 
some kind of L2 hint is almost essential if you don't want to just 
patiently wait for that unsolicited RA, so that you know when it's 
possible to solicit.

> So if we are looking for robustness against bogus L2 "hints" we need to
> consider bogus "link up" hints as an important example in that category.

Yes, definitely.  We need to define what the hints mean and to what 
extent we can infer things from them.  Not just L2 hints, either. 
Reception of an RA with a prefix that we haven't seen before is a hint 
that we have moved to a new network, but doesn't guarantee it. It might 
just be from another router on the network we are already attached to.

Brett.