On Aug 10, 2010, at 9:28 AM, Michael Richardson wrote:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> 
>>>>>> "Carsten" == Carsten Bormann <[email protected]> writes:
>    Carsten> On Aug 10, 2010, at 14:57, Michael Richardson wrote:
> 
>>> The only case where there is a problem is when there is a packet that
>>> arrives from the outside, to a ROLL "border router" (we don't have such
>>> a term in ROLL. The ROLL term would be grounded DODAG root).
> 
>    Carsten> I haven't yet quite found out whether RPL tries to be
>    Carsten> useful for networks with hosts, but if it does, the same
>    Carsten> kind if problem would occur when a host sets the flow
>    Carsten> label. 

...


> 
> For ignorant applications, the OS may well be able to set the
> instanceID.  In other cases, an instanceID of zero may be most correct.
> 
> There may be some cases where there are ordinary hosts which are on a
> network which is not RPL, but is connected to an RPL network. 
> (The group does not have any consensus if these will be layer-2 bridged,
> or layer-3 routed yet).  In these cases the application either will be
> aware, or won't care.

I feel like we're running in circles, in part due to different expectations of 
how RPL will be used.

It's clear that in proprietary or vertically integrated networks running RPL, 
it's possible to state how controllers/other nodes set the flow label in order 
to play nice with RPL. E.g., if my home automation network receives commands 
from a special device (which itself might have a web/SOAP/etc. interface), then 
that special device can set the flow label. This isn't the problem case.

The problem case for the use of the flow label is when arbitrary Internet hosts 
(e.g., my laptop) want to communicate with devices running RPL. E.g., I have a 
home RPL network and want to directly talk to a RPL device from my laptop. 
Requiring the packet source to be aware of flow labels and how to set them 
breaks the whole idea of Internet connectivity: suddenly an end host needs to 
be aware of what routing protocol is used to reach the other host, as well as 
configuration parameters of that host. I need to be able to SSH to the 
RPL-connected device without requiring a new ssh program that uses ioctls to 
set flow labels. We can't say that only OSes which set the flow label to 0 (or 
some other value) can be used. Nor can we expect to change how OSes set the 
flow label.

So the problem is that a packet originating from outside the RPL network, 
destined to a RPL node, may have an arbitrary flow label. The RPL border router 
needs to handle the packet in an effective way.  Whatever solution we come up 
with needs to handle this effectively if the Internet is going to support 
connectivity to LLN devices. It's critical that the solution not make 
assumptions about how the flow labels are set or what the behavior of the 
non-RPL end host is. We want Windows, Linux, OS X, Windows Mobile, Android, 
Symbian, and other current Internet devices to be able to talk with LLNs.

Phil
--------------------------------------------------------------------
IETF IPv6 working group mailing list
[email protected]
Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------

Reply via email to