Hi Erik

> I think the terminology and use is more confused than that.

[Pascal] terminology is a actually a incredibly hard problem because all
words are so overloaded already, and differently for different persons
> 
> The descriptions I've seen of the use of ROLL instances talk of
differentiating
> things for the low delay vs. high throughput packets.
> This used to be called "ToS routing" back when OSPF supported it (I
think it
> was removed because nobody used it.) One could envision something
similar
> for ROLL using the Traffic Class field in the data packets, and
building a
> separate tree for each traffic class.
> 
> But Pascal referred to OSPFv3 instances, which is merely a local
identifier on
> one link to enable multiple OSPF ships in the night sharing e.g., one
Ethernet.
> Thus that instance ID does not span multiple router hops. That has
similar
> effect to having the OSPF control plane run on separate VLANs on that
> Ethernet.

[Pascal] RPL instances are exactly that: ship in the night instances
operating on the same subnet, and that's why we pick the term.
Obviously we extended the concept to a multilink subnet, but that's the
topology we have to live with. And that's the closest abstraction we had
that we could extend to the route over world. 
You'll note that RPL instances, like OSPFv3 instances, are often but not
necessarily used to map traffic classes. They are used to map flows
which can have an arbitrary application meaning.
One of the results of having an instance is that the edge router can
have policies as to where a given instance is redistributed. For
instance, VPN(s) serving metering companies.

> I have no problem keeping the first RFC of a protocol as simple as
possible. If
> the protocol gets widely deployed one can consider extending it later.

[Pascal] RPL describes the operation within one instance. For that
reason, the concept does not make the spec any more complex.
As opposed to ND-09, the spec stands on its own 2 feet with instance 0.
The instance management for other instances is left out.

Cheers,

Pascal
_______________________________________________
6lowpan mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/6lowpan

Reply via email to