Hi Zach,

I agree with some of what you say but not all of it. See inline comments. In a nutshell:

   * We need ND for bootstrapping hosts and routers and to maintain
     hosts. I think nd-09 supports this well
   * Hosts do not need to know anything about routing protocol used in
     the wider network
   * There is nothing wrong with using RPL's (or any other routing
     protocol) implicit multicast mechanism for disseminating lowpan
     wide information

Robert

Robert Cragie (Pacific Gas & Electric)

Gridmerge Ltd.
89 Greenfield Crescent,
Wakefield, WF4 4WA, UK
+44 (0) 1924 910888
http://www.gridmerge.com <http://www.gridmerge.com/>


On 05/05/2010 4:42 PM, Zach Shelby wrote:
Hi,

On May 5, 2010, at 17:18 , Mathilde Durvy (mdurvy) wrote:

Hi Pascal,

I fully agree with you.
In my opinion anything spanning multiple IP hops should not be done by ND.
That logic doesn't work anymore with route-over LoWPANs. As the prefixes are 
spanning the whole LoWPAN there is a natural need for ND to help with that.
<RCC>(1) I repeat - classic ND has no concept over this, so there isn't a natural need for ND to help with this. There has been a choice to add in a mechanism but that mechanism is a 'multicast' mechanism which implicitly has to define propagation. Propagation is not the domain of ND but is the domain of the associated routing protocol</RCC>
  In the case of RPL, the WG may decide to disseminate some things such as 
prefix information between routers in a LoWPAN using the routing protocol. 
6lowpan-nd-09 states clearly that you may do that, so nothing stopping RPL 
there. What is this argument about? You are complaining about something that is 
not a problem?
<RCC>The way it is worded in nd-09 is fine with me, so I have no argument :-)</RCC>
Now RPL isn't the only routing protocol.... 6lowpan-nd-09 provides an optional 
mechanism of using RS/RA for disseminating prefix and context information 
between all routers in a LoWPAN that is routing protocol independent. 
Regardless of RPL, this will surely come in handy and some routing protocols 
may even specify to use this mechanism.
<RCC>See comment (1) above</RCC>
Let me remind you of how RPL started by the way. In the beginning the idea was 
to piggyback RPL information on ND traffic as there were already similar flows. 
Eventually the WG decided to give RPL its own messages instead of piggybacking 
as ND options.
<RCC>Piggybacking RPL options on ND was never going to work as it is back-to-front, same as comment (1) above.<RCC>
Now it has gone to the extreme of RPL being the one and only protocol and 
everything must be carried on that.... Quite a change! Next we could piggyback 
DHCPv6 on RPL, use RPL for DAD, and what the heck, let's go for DNS too... 
Starts to sound like a shopping-TV ad for a super-vegetable-processing-miracle 
doesn't it?
<RCC>What is 'it' ("Now it has gone..."). RPL provides a natural propagation mechanism to all routers so please tell me what is wrong with utilising that to disseminate network-wide information?</RCC>

On a related topic....
6lowpan network have the particularity that you cannot use on-link prefix
due to the non-transitivity of the wireless links. This means we need to
tell routers how to reach neighboring IPv6 hosts. So essentially 6lowpan-ND
is using a registration mechanism to establish a "route" between the router
and the host.
It is actually letting the host and router know about each other (router 
discovery), their reachability (NUD) and their L2 addresses (address 
resolution). These are all standard features of ND.
<RCC>I agree here - we need ND for host - router interaction and indeed for bootstrapping a router onto the lowpan</RCC>
It is not clear to me whether this is the role of ND or of the routing
protocol. I think it could actually be both.
Hence the questions:
- Are IPv6 hosts possible in a 6lowpan network where the RPL protocol is
used?
Better be, or you just broke an important model of IPv6. I would say it MUST be 
possible for hosts to attach to a LoWPAN running RPL (and stay blissfully 
ignorant of RPL).
<RCC>I agree too</RCC>
- Should IPv6 hosts be part of a RPL topology (as leaf node) or should IPv6
hosts use the 6lowpan-ND host-router spec?
ND is clearly the standard host-router interface regardless of IPv6 or 6LoWPAN. 
Forcing hosts to know anything about RPL would be insane...
<RCC>I agree. Hosts should not need to know anything about which routing protocol is in use in the wider lowpan</RCC>
Zach

Best,
Mathilde

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of
Pascal Thubert (pthubert)
Sent: jeudi, 29. avril 2010 09:01
To: [email protected]; Richard Kelsey
Cc: [email protected]; [email protected]
Subject: Re: [Roll] how does a node get an IP address

Hi Zach:

I have yet to review the new ND-09 but my guts tell me that it is the wrong
place to do the job. Router to router is usually routing protocol land and
ND is definitely not a routing protocol.

The main question is how long can  a router advertise a prefix, and the
answer is, as long as it is in the same subnet of an authoritative router
that owns the prefix.
Asserting the continuous reachability of the authoritative router is a
routing protocol problem. Maintaining a subnet together is the job for a new
form of Gateway Protocol, a Subnet Gateway Protocol RPL is just that.

Let see:

- Propagating the RA content is not an ND intrinsic  problem, it only comes
with route over. And route over comes with a routing protocol.
- the route over protocol should be able to tie the route over subnetwork
together so it is a SGP.

So why can't we just say in 6LoWPAN ND that you for those who use it in
route over we expect an SGP to tie the route over subnetwork together and
that the SGP should transport the RA content, maintaining the validity with
the reachability of the authoritative router? I can write that text if you
wish.

It seems that we have a reasonable consensus in this thread to do exactly
that in RPL anyway...

Pascal

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf
Of
[email protected]
Sent: Tuesday, April 27, 2010 10:36 PM
To: Richard Kelsey
Cc: [email protected]
Subject: Re: [Roll] how does a node get an IP address

Hi Everyone,

Let me jump into this thread - just to make things more interesting
;-) First, I
recommend everyone goes and reads 6lowpan-nd-09 which was submitted
today. When it comes to ND, you need to separate two interfaces.

1. The host-router interface

Hosts know absolutely nothing about RPL (nor should they). Thus in
this case
ND* does the job, and RS/RA is used for obtaining a prefix and
initializing its
addresses. I think some people in the thread are referring to this.

2. The router-router interface

As in RFC4861, in 6lowpan-nd-09 routers have more flexibility than
hosts in
how they obtain prefix information (among other things). nd-09 does
include
an optional technique for an authorative border router to disseminate
PIOs
and CIOs (Context Information Options) between the border router and
all
routers in the LoWPAN using RAs. It is actually a decent mechanism and
improved over early versions. The draft clearly states that it is
optional as a
routing algorithm may already do this. So Pascal is correct in that
respect. I
haven't followed the thread well enough to have an opinion if RPL
should do
that.

Routers will also find other features of 6lowpan-nd-09 useful, for
example
during initial bootstrapping, to maintain their default router and
neighbor
caches, avoid the need for address resolution, and to perform NUD. The
draft (tries to) clearly state when features are required or optional
for a
router.

Zach


From: Michael Richardson<[email protected]>
Date: Tue, 27 Apr 2010 10:38:47 -0400

"Richard" == Richard Kelsey<[email protected]>
writes:
Date: Tue, 27 Apr 2010 14:18:32 +0200 From: "Pascal Thubert
(pthubert)"<[email protected]>

The question here is that the authoritative routers need to
disseminate the PIO (and the RIO) to all routers in the
subnet.
    Richard>  How do other routing protocols (OSPF, IS-IS, AODV,
OLSR)
I can only speak for OSPF and ISIS.
Neither deal with multi-hop subnets or with any kind of address
assignment.
Why should RPL be any different?  Yes, it will be run on multi-hop
subnets, but I still do not see how this affects the routing.

Both were written when multicast was very new.
I am not sure how RPL's handling of multicast matters here.
While RPL is required to route multi-hop multicasts, ND uses
link-local multicasts, which do not require routing.

Richard>  I understand that multi-hop subnets are a problem for ND,
Richard>  but I don't see how the routing protocol is affected.

RPL either requires 6lowpan, or it doesn't.
RPL should work fine with ordinary ND.  Why would it require
6lowpan?
If it doesn't, then it has to provide for ND to work, or for
another
protocol to replace it.
ND works fine, using link-local, one-hop multicasts.  RPL need not
be
involved.

If someone wants to run RPL on a node that uses neither ordinary ND
or
6lowpan's version, then they will need some third variety of ND.  I
do
not see why this is an issue for RPL to address.  It seems quite out
of scope.

                              -Richard Kelsey
_______________________________________________
Roll mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/roll


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

Attachment: smime.p7s
Description: S/MIME Cryptographic Signature

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

Reply via email to