Re: [Roll] Iotdir telechat review of draft-ietf-roll-useofrplinfo-40

["Pascal Thubert (pthubert)" <[e-mail]>]

Hello Mališa

Many thanks for your review!

I'm not the editor but I'd like to refine some points here

> 
> Section 1:
> 
> > Since some of the uses cases here described, use IPv6-in-IPv6 encapsulation.
> It MUST take in consideration, when encapsulation is applied, the RFC6040
> [RFC6040], which defines how the explicit congestion notification (ECN) field of
> the IP header should be constructed on entry to and exit from any
> IPV6-in-IPV6 tunnel. - Please clarify the sentence. Consider whether it is
> appropriate to have a normative MUST here.

That's a good point, but then all the best practice of encaps also applies, not sure we need/want to list them all. 
I'm afraid that if we start listing some we miss others and get in an endless discussion at IESG, e.g., fragmentation.

The reverse angle could be the MTU discussion. 
If we use RFC 8138 then the IP in IP does not really steal from the 6LoWPAN MTU of 1280. 
If not it does. So on paper the IP in IP encapsulator should fragment at the IP layer. This might happen in particular in non-storing mode with a routing header that causes a larger encapsulation.
But as it goes, even though the MTU is 1280, the fragmentation allows more (2048 with RFC 4944, no explicit limit with https://tools.ietf.org/html/draft-ietf-6lo-fragment-recovery-21).

So it could be good to add text that says:
- best practice IPv6 encapsulation applies
	- mention RFC6040
- there are exception
	- no IP layer MTU discovery (because MTU is 1280)
	- though MTU is 1280, no IP layer fragmentation when the encapsulated packet is over that (because fragmentation copes with it above 1280)
	- The recomposition buffer SHOULD be large enough. We could RECOMMEND 2048 bytes.


> Section 4.2:
> > The non-storing mode case does not require the type change from 0x63
> > to 0x23,
> as the root can always create the right packet.  The type change does not
> adversely affect the non-storing case. - It is not clear what RPI option type
> should non-storing networks use. A pointer to the discussion in Section 4.3
> would be useful.
> 

Note that adding/removing an RPI on the way (e.g., at the root) always involves en/decapsulations (ask 6MAN!). 
I'm not clear what difference you make there?

The problem is for packets from the LLN node to the internet. This is the same for storing and non-storing.
0x23 allows us to not remove The RPI thus to not encapsulate at the LLN node. Maybe that can be clarified.

We deprecate 0x63 globally and want it replaced by 0x23 in all cases. Maybe that can be clarified too. 


> Section 4.4:
> 
> > A node that is decompressing this header MUST decompress using the RPI
> > Option
> Type that is currently active: that is, a choice between 0x23 (new) and 0x63
> (old). The node will know which to use based upon the presence of the flag in
> the DODAG Configuration option defined in Section 4.3. E.g.  If the network is
> in 0x23 mode (by DIO option), then it should be decompressed to 0x23. - If my
> understanding is correct, this means that in order to decompress data plane
> packets, a node first needs to remember the option type mode the network is
> operating in, advertised in DIOs. Consequently, decompression is not possible
> before at least one DIO is received.
 
True. At that point the node does not even know it is in a RPL network. 
Note that the decompression generally happens at the destination of the outer header which removes the RPI if it is a router and ignores it if it is the final destination. So it does not really matter.
But just in case we can mention that the default is 0x23.


> Section 6:
> 
> > The RPI MUST be present in every single RPL data packet.
> - How is the normative text here appropriate at this point? Is this not
> redundant with RFC6553?

Agreed to lowercase it, and as you point out all the uppercase which echo an existing RFC

> Section 8:
> 
> > The root always have to encapuslate on the way down
> - It is not clear how come does root need to always encapsulate on the way
> down. In the basic case of root to RAL communication, IPv6-in-IPv6 is marked
> as “No”. Please clarify.

Need to clarify that this is for routed traffic not self-generated traffic

> Section 8.2.1:
> 
> - A sentence stating how does RAL recognize that the packet is destined for the
> Internet would be useful.

Note that that it does not need to. With this spec, the RAL may do the same thing whether the destination is a RAL, a RUL, or the Internet.
But yes I agree; we could say that if the RAL knows what the RPL domain is, it may encapsulate to the root when the destination is not in that domain.


Many thanks again!

Take care

Pascal