Re: [Roll] [manet] Understanding RPL [was: Working group re-charter ideas]

[Juliusz Chroboczek <[e-mail]>]

Thanks a lot for the lengthy explanation, Pascal, I feel that I understand
RPL somewhat better now.

>> My perhaps mistaken understanding is that RPL uses a more primitive
>> version of DSDV's feasibility condition

> That's all this discussion around that value L

Ah, I see.  Section 8.2.2.4.3.  (I'm not apologising for missing it -- it
is well hidden.)

Yes, that looks like DUAL-feasibility, but weakened so that loops of
diameter DAGMaxRankIncrease or less are allowed, right?  Why the
weakening?  Is that to compensate for the fact that you don't have
a mechanism to resolve blackholes in a timely manner?

> A node is allowed to join any DODAG Version that it has never been a
> prior member of without any restrictions, but if the node has been a
> prior member of the DODAG Version, then it must continue to observe
> the rule that it may not advertise a Rank higher than
> L+DAGMaxRankIncrease at any point in the life of the DODAG Version.

Yeah, that's pretty clear.  The same is true of Babel.

> One interesting difference with Babel is that RPL can route to all nodes
> with a single DODAG to a single root, using MOP 2.

Yes, I've noticed this mechanism before.  I think that's a smart mechanism
(I much prefer it to MOP 1), and I trust you've evaluated its performance
carefully, at least for some OF.  However, that's definitely not something
I'd ever consider adding to Babel -- I have a very strong desire to keep
the base protocol small, and to make sure that all extended variants can
interoperate as long as they implement the base protocol correctly.

You see -- once again, we're aiming at two different application spaces.
You're thinking of sensor networks, where it is reasonable to require that
all nodes in a given network implement MOP 2.  Babel is aimed at home and
hybrid networks (cash-starved networks with some meshy bits), which tend
to grow organically, with routers that never get reflashed, and therefore
contain different variants of the protocol.

> Another interesting difference is that if we have 2 roots, RPL
> partitions the network in 2 DODAGs.

Sorry if I'm being slow, but I don't see how that can be true.  If you have

   R1    R2
   /\
  /  \
 A    B

and there is instability in the network, so that simultaneously A switches
to B as parent, and B switches to R2 while its retraction ("poison") gets
lost, what prevents A from having B as parent while believing it's still in
R1's DODAG?

> Local repair is done by either poisoning or detaching, sections
> 8.2.2.5.

I don't see how that can be implemented.

Suppose A is B's parent.  A advertises INFINITE_RANK.  A's retraction
("poison") is lost.  B has no reason to change parents, so B has a parent
with infinite rank.  Oops, B is just violating the MUST NOT in 8.2.2.5.2.

Are you assuming the network is reliable?  Or am I missing some mechanism
that prevents that from happening?

> Or 8.2.2.6.

That's just a variant of the above but advertising DIO instead of
a retraction, right?

> On the side, poisoning is really inheriting from IGRP,

It's been a long time since I looked at IGRP, but I seem to recall that it
avoids the issue I describe above by imposing a silent time after a metric
increase.  This causes temporary blackholes, of course, of up to 3 minutes
if memory serves.

> We voluntarily left out the trigger for a new iteration, because that
> can be largely outside the scope of the routing protocol,

I disagree with that, at least in the space for which Babel is aiming.

> All in all I do not buy the different spaces. RPL was explicitly
> designed for multiple spaces because, by design, we left out of RPL and
> into the OF everything that's specific.

Well, the space Babel aims for requires that implementations from different
vendors interoperate.  That requires putting a provably complete set of
"must implement" mechanisms in the base spec.  I don't believe that RPL
has that property.

> Do I get an invite?

Sure, I'll try to remember to send you a note when I next give a seminar.

-- Juliusz