Elwyn
> Elwyn Davies wrote:
> [...]
>
> 3. The e2e mutability issue has been done to death in DiffServ for
> code points and it sounds like the whole argument is being trotted out
> again. The one and only fundamental point is that if the flow label
> is mutable, information is generally lost when it is changed after
> original setting and can never be restored unless there is some
> additional agency available to regenerate the information (entropy
> argument). If the label is changed in a way that does not lose
> information then it doesn't matter; likewise, if you don't care about
> the loss of information then go ahead and change it.
>
More explicitly, the information can be a combination of labels and
state in
each router, ala MPLS. Labels can change from one node to another, but
the
state in which node, links the input label value to the output label
value,
without affecting the information being transmitted.
> 4. The flow label could clearly (also) be used as a signalling method
> to help out with inter-domain MPLS issues when you need to cross
> exchange points where the whole MPLS header has to be removed.
>
I am not sure about the practical value of this.
If two MPLS domains are logically disjoint, I think that there are
practical
reasons - they are not physically adjacent, or if they are, the logical
separation is desired, which seems to make more sense with a new MPLS
classification at the downstream domain border.
> I concur fully with Brian Carpenter's points on orthogonality of
> DiffServ and MPLS.
[...]
>
> Regards,
> Elwyn Davies
>
I think the orthogonality is relative.
MPLS labels may carry Diffserv, and Intserv information, and may be
derived
based on Diffserv, and Intserv information. I would not necessarily call
that
orthogonal.
Alex
S/MIME Cryptographic Signature
