Hi Mark,
Mark K. Thompson wrote:
Hi,
On 9 Aug 2005, at 11:53, Arifumi Matsumoto wrote:
On 2005/08/09, at 5:40, Mark K. Thompson wrote:
the lack of field definition for labels has seen different OSes use
different datatypes for the label, from string through
stringified-integer to integer. Any cross-platform policy
specification protocol would need to cater for this inconsistency
By our protocol, we intend to distribute an address selection policy,
of course, but the values in this option aren't absolute but relative.
When an option includes the following, for example:
Prefix Precedence Label
2001:db8::/48 100 10
3ffe:db8::/48 20 10
2002:1:2::/48 10 5
The Label values just indicate that the first line and the second line
has the same label and the third line doesn't have the same label as
any other lines.
This is the case with Precedence value. This option indicates the order
of precedence.
So, if a host has its policy table configured manually and has the label
value 10, the distributed policy's label value should be changed to
another un-used value or string.
About Precedence, the values in distributed option can be changed
as far as the order isn't changed.
So, this just seems to me an issue of implementation on DHCPv6 option
receiver. It may be better to clarify these implementation
considerations
in our draft or in another draft.
This strikes me as quite non-deterministic and I wonder whether
administrators would be happy having pushed-policy being interpreted
differently on different nodes depending on their individual state at
the point where the policy was received.
For example, I can envisage scenarios where administrators want to
assert precise control over the contents of the policy table
exclusively over any local settings - and that policy may be node-
specific as determined by DUID or some Vendor option in DHC. (Assuming
that a DHC approach was deemed suitable)
I can't see this as particularly useful, and there are better ways
(for example PANA address assignment or VPNs) to tie this to the
access control state of the host.
Otherwise, the control is superficial. Hosts or applications
can ignore or reject the advice of the server, and configure what
they like.
Greg
--------------------------------------------------------------------
IETF IPv6 working group mailing list
ipv6@ietf.org
Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------
