RE: [DNA] RE: Review of draft-krishnan-dna-simple-01.txt

[Bernard Aboba <Bernard_Aboba@hotmail.com>]

  > > > 4) When 2 & 3 give conflicting answers, 3's decision is definite.
  > >
  > >  [BA] Correct -- except for DHCPv6 case, where 2) may confirm
  > >  the validity of a previously assigned address, but 3) may indicate
  > >  the need to send a DHCPv6 request.   As before if the DHCPv6
  > >  assigned address conflicts with the address determined in 2),
  > >  then DHCPv6 wins.

[JC] However, from your remarks below, it seems that sometimes NS/ NA based
decision overrules RA based decision.

[BA] I think that an RA from a router will over-ride an NA from that same
router, with respect to a given address.  But an RA from another router will
not. 

>  If a NUD exchange suceeds, the host should merely assume
>  that the information it got from that particular router is still valid,
not
>  that
>  it has confirmed *all* the information from *all* the routers.   So if it
>  assigned a still-valid address based on a particular prefix announcement,
>  it need only confirm reachability to the router that sent that prefix
>  announcement.  Receiving an RA with no known prefix from some other
>  router is immaterial.

[JC] ok. However this necessitates hosts to maintain the state of pairs, (a
prefix, a router which advertise the prefix). Also if an RA with no
known prefix arrives before a solicited NA, I assume the host
immediately decides a link change, instead of waiting for the NA.

[BA] The way I think of it is that the host needs to determine
whether its existing addresses are valid.  It can determine that based
on receipt of an NS from a router, or receipt of RAs.   

Let us assume that formerly router A announced a prefix from which the
host formed address A.  Now with Simple DNA, the host tries to determine
if that address is still valid.  In response to an RS, it receives an RA
from that same router with no known prefix.  It can now conclude that 
address A is not valid, under the assumption that the prefix should have
been included if the router was still advertising it. 

However, the host could also have formed another address B based on an
announcement from router B.  With Simple DNA, the host sends a unicast NS
to router B which responds with a solicited NA.  The host can now assume
that
address B is still valid -- unless it receives an RA from router B
that is missing the prefix from which that address was derived.

[JC] I have difficulty discerning the above mechanism from the draft.
That's why I first tried to clarify DNA operation lest we should
discuss over irrelevant items.

[BA] You are correct that the current draft does not describe how 
Simple DNA needs to operate.  As you state, for simple DNAv6 the 
host needs to maintain lists of valid addresses and the 
corresponding routers from which they obtained those addresses.  
This is how DNAv4 works. 

>  And of course, if that same router updates its
>  prefix list,
>  then the previously cached DNA configuration information is invalidated.

[JC] ND allows a router to omit some prefixes when it advertises an RA. So
from a prefix list in an RA, a host has difficulty discerning whether
a missing prefix is invalidated or simply omitted.

[BA] Right -- but if the router wants to deprecate a previously announced
prefix, it does have to announce it, no?  We are assuming that the host
only uses simple DNAv6 to re-confirm addresses that would otherwise be
still valid (e.g. addresses whose DHCPv6 lease has not yet expired, or
whose prefix is still valid).  If the address is invalid, then the host
should not be using simple DNAv6. 

>  As an example, if a host previously recieved an RA with no known prefix,
>  and as a result got a valid address assigned via DHCPv6, it should be
able
>  to confirm the validity of that address based on a NUD exchange with the
>  router (while doing a DHCPv6 configuration exchange in the background).

[JC] This is not clear. With which router a host should perform NUD to
validate a DHCP based address? No router has specific relation with
the DHCP based address. Any router will do? If that's the case, why
can't we use that criterion for non-DHCP based address.

[BA]  To validate a DHCPv6-based address, the host should perform NUD
to the router which lead it to believe that DHCPv6 was supported on
the link.  If it receives an NA, it can assume that the DHCPv6
address is still valid, assuming that the original lease has not yet
expired.  Of course, if it receives an RA from that router indicating
that conditions have changed (e.g. now DHCPv6 is no longer supported),
or receives a response from the DHCPv6 server indicating that another
address is being assigned, then the host's configuration should change.

[JC] Now I come to wonder why RA based decision is assumed definite. Maybe
it's better to make NS/NA based decision takes precedents over RA
based decision.

[BA] Receipt of an RA is additional information which may invalidate an
assumption made as a result of receipt of an NA.  For example, the RA
could indicate different prefixes, or a change in address assignment
mechanism (stateless to stateful or the other way around). 

In any case, I think it is important that we write all of this down in the
Simple DNAv6 draft so that we can evaluate it and understand how well it
works and under what conditions it might not work.   I had sent my comments
on the draft in a previous mail.