[Crossposted to dhcwg even though I'm not on that list, as people
there may be able to add some useful insights.]
On 24-mei-2005, at 16:45, Thomas Narten wrote:
I wonder if there is key question here that the community has simply
not agreed on (yet), and that that question is at the heart of all
this "confusion".
Does the community feel that operators need RA bits that
control/indicate whether a client is to invoke DHC? That is, is there
a need for the sys admin to signal to client whether DHC is to be
invoked?
By a strange coincidence, I spent most of the day yesterday taking
several DHCPv6 implementations for a test drive, for reasons
unrelated to this discussion.
These implementations are: KAME DHCP6, the unnamed Linux fork of the
KAME implementation at http://dhcpv6.sourceforge.net/ and the Cisco
IOS implemenation.
Conclusion: the Cisco implementation is incomplete (no address
assignment) and the other two are too immature to be of much use.
I'm impressed with the prefix delegation functionality, but as
before, the prospect of running such a complex protocol just to
optain a domain search list and some DNS resolvers makes me very
uncomfortable.
All of this, coupled with the fact that, AFAIK, no OS implements
DHCPv6, means that there is essentially zero experience with DHCPv6
in the operational community. This means that at this time, there is
little point in asking the operational community what should happen
with the M and O bits.
In other words: this is not the time declare the bits useless and
remove them.
Second, is it important that such a signal be honored by clients?
(That is, if clients end up mostly ignoring the flags, does their
presence become useless?)
Depends. There are three possible ways this can play out:
- the DNS resolver issue is solved in a way that doesn't requite
DHCP, so most people don't don't run DHCPv6 at all, others run it all
the time -> no bits necessary
- the DNS resolver issue isn't solved in another way, so everyone
runs some form of DHCPv6 all the time -> no bits necessary
- DHCP provides benefits but some people are reluctant to use it ->
helpful to know whether to bother running DHCPv6
For example, should the sys admin be able to say "do not run DHC,
doing so wastes local resources and won't get you any config info"?
(And should that be honored by clients?)
I think it's good to recognize that in the past, there have often
been security issues with non-text based UDP protocols, so knowing
there is no need to run DHCP and then not running it would be good
security.
Fundamentally, it is only the access network that has knowledge of
whether running DHC is useful. Thus, by default, clients (arguably)
can't know whether running DHC is useful, so by default they shold
invoke DHC (unless the sys admin signals "don't invoke DHC").
Or (switching the argument), by default, client should not invoke DHC,
unless the local sys admin indicates doing so is appropriate.
There isn't really a difference here, except for what happens when
there are no RAs.
I would be interested in hearing viewpoints on the usefulness of
running DHCPv6 even though the hints indicate that there is no need to.
--------------------------------------------------------------------
IETF IPv6 working group mailing list
ipv6@ietf.org
Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------