Le 20/10/2012 18:36, Mikael Abrahamsson a écrit :
On Sat, 20 Oct 2012, Alexandru Petrescu wrote:
Right, parts of ND are handled in kernel in most OSes. But one
key part that would need to be modified is RA and that is
userspace. In linux that means mainly radvd, and curiously enough
that lacks RS which is mostly kernel.
Sending of RA is userspace, reception of RA is kernel.
Right, and that may make it an issue. Our implementation of reception
RA is userspace. I think it does not interfere with what the kernel
first does with that RA. This should be ensured.
So, instead of preferring DHCP one may bring RS out of kernel into
radvd (a so called 'rsadvd'). At that point it's relatively easier
to do Prefix Delegation with ND than with DHCP,
implementation-wise, not least because code is smaller.
Why is it easier? Also, I'd imagine there are already
implementations of DHCPv6-PD available under quite permissive
licenses, instead of what you're proposing which requires everybody
to develop new code.
Well, not everybody should develop new code. Just download some :-)
But there is indeed a point about compatibility with existing ND
implementations. There may be a need to make sure that the proposed PD
extension implementation does not break other computers' ND stacks when
faced with it. I think this would be enough, no?
Alex
Where does PD belong - to DHCP or ND is a long discussion and I'll
be happy to take part if it happens.
For me, this is that discussion.
--------------------------------------------------------------------
IETF IPv6 working group mailing list
ipv6@ietf.org
Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------
