On Fri, 21 Jan 2005, Brian Haberman wrote:
    This Last Call ended yesterday with no comments.  At a minimum,
I would like to hear from those people who commented previously
whether or not this version satisfies their concerns.

Hi,

I took a quick look at -03, with comments below. I think the main oDAD functionality is OK, but the draft likely needs a revision to make a few textual corrections.

substantial
-----------

Sorry that I did not catch this earlier, but I think the analysis in
Appendix A on collision probability is not quite accurate.

[SOTO] describes the collision of a random interface identifier.  For that,
looking at 62 bits is OK.

What we are mostly concerned with are:
 - manual configuration
 - stateless address autoconfiguration
 - RFC3041 addresses

Now, we cannot really analyze manual configuration mismatches, but we can
analyze the probability of collision of SLAAC addresses.  For RFC3041, the
current calculations would apply, but we should calculate the maximal
feasible probability as it is larger with SLAAC addresses.

And for that, you'll have to use a different instead of 2^62; you could take
for example Ethernet, and analyze 2^48.  Text also needs to be massaged a
bit to state which probability you are actually calculating. (Unfortunately,
as the L2 address lengths are variable, you can't do much better
calculations than this..)

I don't think this changes the results in any meaningful way (the
probabilities are about 16K larger), but this should probably still be
changed.

semi-substantial
----------------

   * Nodes implementing Optimistic DAD SHOULD additionally implement
        Secure Neighbor Discovery [SEND].

==> I agree with Jinmei's concern.  SEND should be a normative reference if
it's a SHOULD.

   * (modifies 5.5) A host MAY choose to configure a new address as an
        Optimistic Address.  A host which does not know the SLLAO of its
        router SHOULD NOT configure a new address as Optimistic.  A
        router MUST NOT configure an Optimistic Address.

==> Why MUST NOT for routers?  It seems too strong.  Sooner or later (if not
already) folks deploying Mobile Routers (for example) are going to want
to use this for their upstream interface, at least.

Either you should justify clearly somewhere why it's technically unfeasible
to use oDAD on routers, or weaken the text a bit, e.g., to a SHOULD NOT.

editorial
---------

   The existing IPv6 address configuration mechanisms provide adequate
   collision detection mechanisms for the static hosts they were
   designed for.

==> s/static/fixed/ ?

  SAA - Stateless Address Autoconfiguration.  The process described in
        [RFC2462]

==> "SAA" is ambiguous, because it could as well be Stateful AA.  I suggest
using either SLAC or SLAAC (both have been used).

4.2 Collision case

==> this section does not explicitly state at which point the ON
unconfigures the address.  Should it say that?

   When considering collision probability, the Birthday Paradox is
   generally mentioned.  When randomly selecting k values from n
   possiblities, the probability of two values being the same is:

==> s/possib/possibi/

   When considering the effect of collisions on a individual nodes, we

==> "a invididual nodes" ?

--------------------------------------------------------------------
IETF IPv6 working group mailing list
[email protected]
Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------

Reply via email to