Re: [Roll] [Anima] ACP RPL profile and dynamics

[Hendrik Mahrt <[e-mail]>]

On 9/9/22 01:33, Toerless Eckert wrote:
> On Thu, Sep 08, 2022 at 12:44:05PM +0200, Hendrik Mahrt wrote:
>>> On the other hand, a more typical ISP situation is there is a router with
>>> three or four WAN links, each of which is a p2p ethernet.  In that case,
>>> there is really only one peer on each link, and it makes no sense not to have
>>> a tunnel up on every interface.
>>>
>>
>> I'm not quite sure how the type of media or its capability to broadcast
>> interacts with RPL parent selection. The way I understand ACP, IPSec
>> tunnels are established with all link neighbors of the same ACP domain.
>> This is done prior to RPL coming into action.
> 
> Right. Just like any L2 security would happen before RPL exchanges messages.
> 
>> It is also necessary to
>> exchange RPL ranks with all neighbors. How else would a node determine
>> its parent(s)? I guess afterwards tunnels to neighbors that are neither
>> parent nor child of a node could be closed again, yes.
> 
> Which IMHO would raise the problem of then being unable to discover all
> RPL changes from a (closed) RPL neighbor, unless there are clear RPL procedures
> that indicate whenever a RPL neighbor needs to actively reach out to another
> RPL neighbor again (hence being able to have signals to re-create the ACP
> tunnel).

yes that is how I understand it as well. So, tunnels must be established
between all nodes for RPL signaling. They may be closed but must be 
re-established if RPL requires to reach that neighbor.

> 
>> The wording in ACP Section 6.12.1.7 is "The DODAG version is only
>> incremented under catastrophic events", therefore I was under the
>> impression global repair would only be done in extreme circumstances,
>> and not periodically.
> 
> So my "Root dies" could be seen as a catastrophic event. What then would
> be the least-catastrophic event that could only be handled with periodic
> version increase ?
> 

They way I understand RPL, any form of mobility in the DODAG could lead
to this problem if no periodic global repair is performed. Mobility
either in form of extensive physical mobility of a node or as movement
in the DODAG caused by link failures, where the rank increase is too
high. For example, after a link failure a subtree may only be attached
to the rest of the tree by a link further down below. But I may be
entirely wrong and have missed some other mechanism in the RPL standard,
that allows a node to change its rank beyond the maximum rank increase
in the same DODAG and same version.

I would really appreciate if someone with RPL experience would come to
the rescue here. Also for my other original question about the
greediness of nodes when selecting multiple parents in this special RPL 
profile used in ACP.
I will also read the RPL standard front to back again to see if I missed 
something.

> Cheers
>      Toerless
> 
>>   Hendrik

  Hendrik