Re: [Roll] WG Last Call draft-ietf-roll-trickle-mcast-02

[Robert Cragie <[e-mail]>]

Hi Dario,

Some comments inline. I have also attached some diagrams which hopefully 
help to illustrate some of the thought processes we went through when 
implementing and testing between the ZigBee IP vendors (comments 
welcome). Apologies in advance for the PDF format but ASCII art would 
have been difficult :-)

Robert


On 31/10/2012 11:22 PM, Dario Tedeschi wrote:
> Yes, I'd like to try clarify why link-local scope was suggested for 
> the *outer* header. The reasons were:
>
>  1. Link-local scope is the only scope where the boundaries are well
>     defined (i.e. on the link). Higher scopes are not well-defined and
>     can cover wide domains depending on network configuration and
>     administration.
>
<RCC>I agree and made essentially the same comment re. higher scopes.</RCC>
>
>  1. With a higher scope, there is a chance that non-MPL aware routers
>     may simply forward encapsulated multicast datagrams (MPL HbH
>     option and all). We wouldn't want MPL datagrams to leak outside of
>     an MPL domain.
>
<RCC>Which is why I think the encapsulation rules do need to be pretty 
specific. If link-local encapsulation is always used then providing the 
MPL forwarder rules are clear, the MPL domain is then entirely bounded 
by the MPL forwarders and there is no question regarding address scope 
and administration thereof. This cleanly covers Peter's case as well 
where he wants to forward into another PAN - it would be processed 
internally as an original non-MPL packet and then be "launched" into the 
other PAN using LL encapsulation for that PAN. Using other scopes for 
the outer header would still work but then there is the issue of 
administering the scope. However, this would need to be done in the case 
where no encapsulation is done anyway.</RCC>
>
>  1. A higher scope complicates the forwarding logic that needs to be
>     implemented in an MPL router. The complication comes when a router
>     receives an MPL datagram and needs to figure out whether to
>     decapsulate or not. Granted, the use of an MPL group would
>     mitigate this problem to a degree, but link-local scope would make
>     the decision to decapsulate very obvious and simple.
>
<RCC>I think it would have to be effectively decapsulated at every 
router anyway irrespective of the scope of the outer header to see if it 
needs to be processed - it is the inner header which counts there and 
the comments about multicast groups come into play in that discussion. 
But I agree using link-local makes that decision easy and somewhat 
clearer</RCC>
>
>  1. In conjunction with 3. Link-local scope also makes it easier for
>     an MPL router to determine if the inner multicast address is one
>     that a higher layer (or an app) may be interested in.
>
<RCC>Agree that the rules are clearer for link-local</RCC>
>
> 1.
>
>
> Hopefully I haven't made things more confusing.

<RCC>Perish the thought ;-)</RCC>
>
> - Dario
>
> On 31/10/2012 7:53 AM, Jonathan Hui (johui) wrote:
>> Hi Peter,
>>
>> The current draft does not place any restrictions on the MPL multicast scope.
>>
>> If the LLN border router is an MPL forwarder, it can forward MPL multicast packets between different MPL multicast scope zones.  To be explicit, if the original multicast packet's destination address has link-local scope, the MPL forwarder should not forward the packet again.  If the original multicast packet's destination has a scope larger than the MPL multicast scope, then the MPL forwarder needs to forward the packet to other MPL multicast scope zones (which may or may not involve different interfaces).
>>
>> Does that address your question?
>>
>> --
>> Jonathan Hui
>>
>> On Oct 31, 2012, at 3:54 AM, peter van der Stok<[e-mail]>  wrote:
>>
>>> Hi Jonathan,
>>>
>>> To be absolutely sure: the MPL multicast scope can be link-local, ULA or site-local? meaning the LLN border router can be a MPL forwarder?
>>> In the latter case the LLN border router can forward link-local multicast from one interface to another?
>>>
>>> Greetings,
>>>
>>> peter
>>>
>>> Jonathan Hui (johui) schreef op 2012-10-30 18:27:
>>>> Yes, a goal of the current draft is to support both cases (use of
>>>> IPv6-in-IPv6 encapsulation or not).
>>>>
>>>> The intent is as follows:
>>>> 1) If the source of the multicast packet is within the MPL forwarding
>>>> domain and the destination has a scope equal to or smaller than the
>>>> MPL multicast scope, then no IPv6-in-IPv6 encapsulation is required.
>>>> 2) If the source of the multicast packet is outside the MPL
>>>> forwarding domain or the destination has scope greater than the MPL
>>>> multicast scope, then IPv6-in-IPv6 encapsulation is required.  When
>>>> using IPv6-in-IPv6 encapsulation, then the all MPL forwarders
>>>> multicast address with scope = MPL multicast scope is used as the
>>>> destination address in the outer header.
>>>>
>>>> As mentioned in my other email, IPv6-in-IPv6 encapsulation is
>>>> required if you want to use the IPv6 Destination Address to identify
>>>> MPL forwarding scope zones.
>>>>
>>>> Of course, this brings up Dario's practical point of how to configure
>>>> the MPL multicast scope if we allow that to dynamically change.  He
>>>> proposes a simplifying suggestion of requiring IPv6-in-IPv6
>>>> encapsulation for all non-link-local multicasts.  In other words,
>>>> setting the MPL multicast scope to link-local.
>>>>
>>>> Thoughts?
>>>>
>>>> --
>>>> Jonathan Hui
>>>>
>>>> On Oct 30, 2012, at 4:46 AM, Don Sturek<[e-mail]>  wrote:
>>>>
>>>>> Hi Peter,
>>>>>
>>>>> I still need to read the latest draft so take what I say here with that in
>>>>> mind......
>>>>>
>>>>> I was hoping that we could support not using IP in IP tunneling if the
>>>>> scope of the multicast transmission was only within the multi-link subnet
>>>>> managed by the border router.   I was hoping that only transmission
>>>>> emanating from outside the multi-link subnet, received at the border
>>>>> router, with scope that includes the devices in the multi-link subnet
>>>>> would require IP in IP tunneling (and vice versa in terms of multicasts
>>>>> generated in the multi-link subnet with scope outside).  I haven't yet
>>>>> read the draft carefully to know if this is possible.
>>>>>
>>>>> Don
>>>>>
>>>>>
>>>>> On 10/30/12 1:34 AM, "peter van der Stok"<[e-mail]>  wrote:
>>>>>
>>>>>> Hi Don,
>>>>>>
>>>>>> and more specifically under which conditions. That gives the
>>>>>> possibility to choose the conditions such that the encapsulation is not
>>>>>> needed.
>>>>>>
>>>>>> Don Sturek schreef op 2012-10-29 16:56:
>>>>>>> Hi Peter,
>>>>>>>
>>>>>>> I think your suggested changes to the Trickle Multicast draft point
>>>>>>> out
>>>>>>> why IP in IP tunneling is needed.
>>>>>>>
>>>>>>> Don
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> On 10/29/12 3:44 AM, "peter van der Stok"<[e-mail]>  wrote:
>>>>>>>
>>>>>>>> Dear WG,
>>>>>>>>
>>>>>>>>
>>>>>>>> Attached my suggestions for text modifications including some nits. I
>>>>>>>> used the facilities of word to edit and comment text with traces.
>>>>>>>>
>>>>>>>> When writing text about MC scope and MC domain, I was puzzled by the
>>>>>>>> all MPL forwarders multicast address which removes the possibility to
>>>>>>>> address a given multicast group. We expect multiple (possibly
>>>>>>>> disjunct)
>>>>>>>> MC groups in our wireless networks.
>>>>>>>> Also I failed to understand why encapsulation was necessary once the
>>>>>>>> message was received by the seed.
>>>>>>>> To make it possible to configure the interface with one MC scope I
>>>>>>>> added the possibility to use Unicast-Prefix-based IPv6 Multicast
>>>>>>>> Addresses (RFC 3306).
>>>>>>>>
>>>>>>>>
>>>>>>>> Probably, I overlooked many aspects which make the suggestions
>>>>>>>> impractical, but I hope that the intention is clear.
>>>>>>>>
>>>>>>>> Peter van der Stok
>>>>>>>>
>>>>>>>> Michael Richardson schreef op 2012-10-25 23:30:
>>>>>>>>> I suggest that you propose specific text to the list to modify the
>>>>>>>>> document.
>>>>>>>> _______________________________________________
>>>>>>>> Roll mailing list
>>>>>>>> [e-mail]
>>>>>>>> https://www.ietf.org/mailman/listinfo/roll
>>>>> _______________________________________________
>>>>> Roll mailing list
>>>>> [e-mail]
>>>>> https://www.ietf.org/mailman/listinfo/roll
>> _______________________________________________
>> Roll mailing list
>> [e-mail]
>> https://www.ietf.org/mailman/listinfo/roll
>
>
>
> _______________________________________________
> Roll mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/roll