Hi Julien,
On 10-09-08 07:43 PM, Laganier, Julien wrote:
Thomas Narten wrote:
[...]
"Joel M. Halpern" <j...@joelhalpern.com> writes:
[...]
Now, operators wanted to offer IPv6 service. I hope we think that is a
good thing. For residential, they looked at what they could count on
from the hosts. And some of them concluded that they could not count on
DHCP, so they designed an architecture around SLAAC. In other words,
they ddi what we told them to do.
Only partly true. The proposal is to use SLAAC and RAs in an
environment in which basic assumptions on which SLACC and RAs were
built do not hold.
RAs/SLAAC work very well when RAs can be multicast to *all* nodes on a
link, and *all* nodes receive exactly the same information about
prefixes and SLAAC. I.e, your normal subnet model.
What BBF is proposing to do, is to use RAs/SLAAC where each customer
node (i.e, each node on the access network) recieves *different*
configuration information. This means that multicast model on which
RAs were built doesn't work.
This is hardly a "normal" IPv4 subnet model even. (I.e., the nodes
don't share a common prefix, even though they are in the same
broadcast domain.)
Right, what we essentially have here (and what has been referred to as "N:1 VLAN allocation model") is a point-to-multipoint link model that constitutes a broadcast domain but where there is a desire to advertize different prefixes to different endpoints (CPEs at the multipoint end.)
The 16ng WG has managed to do so something similar over the IEEE 802.16 link
layer that is documented in RFC 5121. In 16ng however, from the perspective of
the edge router, each of branch of the point-to-multipoint 802.16 link layer
appears as a distinct link to the IP layer, thanks to the use of distinct,
per-endpoint, IEEE 802.16 connection IDs and tunneling between the edge router
and the aggregation point. Because there is a different link between the edge
router and the endpoints, the edge router can multicast unsolicited RAs
containing a per-endpoint prefix on each of the links.
In BBF this cannot be done because there is no tunneling between the access
node and the edge router, and thus the edge routers sees only one link.
Maybe it is time to write an "Issues with Per-Endpoint Subnet Link Model on
Point-to-Multipoint Links" that would explains the caveats that exists, ala RFC 4903.
This sounds like a good idea. I think identifying the generic issues
with this kind of usage will be invaluable. I am looking forward to read
and contribute.
Thanks
Suresh
--------------------------------------------------------------------
IETF IPv6 working group mailing list
ipv6@ietf.org
Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------
