Re: [ipwave] Expertise on ND problems on OCB

[Alexandre Petrescu <[e-mail]>]

Le 15/04/2019 à 16:16, Charlie Perkins a écrit :
> Hello Alex and all,
> 
> I have a few comments, although I have only briefly scanned the OCB 
> draft.
> 
> * RFC 8505 isn't just about low power.  More to the point in this 
> discussion, it's about reducing interference by reducing the number 
> of multicasts needed in a multilink subnet.  Please note that I am 
> neither discouraging nor advocating the use of RFC 8505 for 
> v6-over-OCB, but I think it would be better to avoid suggesting that
>  it is only useful for low-power mesh.

I will avoid suggesting RFC8505 is only for low-power mesh.  Ideally I
would like to know which power levels are used in these meshes, and
group them with the power levels in OCB, and maybe call them a range
where that RFC may apply.

> * Even if you did have four cars to run experiments, the results 
> could hardly be considered conclusive.  Simulation can be very 
> helpful in such circumstances.  You know the velocities and the
> range of the signal. You have many many thousands of easily
> accessible real-world topologies, and even the interference
> characteristics.

I agree, maybe both simulation and car use would be conclusive.

> * I made a quick scan of the draft and found this: "All nodes in the
>  subnet MUST be able to communicate directly using their link-local 
> unicast addresses.".  Does this mean that the solution is not 
> applicable for topologies exhibiting the hidden terminal problem?
> Or, does "directly" mean something else?

'Directly', to me, means directly at networking layer (layer 3).  To me,
it also means these link local addresses are present in the src and dst
fields of the first IP header of a packet.

The topologies with the hidden terminal problem: I did not see them in
practice.

The hidden-terminal problem was formulated for WiFi non-OCB.

I do not know how the hidden terminal problem works.  I feel like if I
start looking into it I will grow additional white hairs.  I feel like I
better avoid  it.

I think other people have some experience with the hidden terminal
problem in non-OCB WiFi and 802.15.4.  I am listening to them.

I think nobody has any experience with the hidden terminal problem in
OCB settings.  I have some doubts, but still I think it is almost nobody.

> I think the draft could benefit from a very clear applicability 
> statement, specifically including the projected number of vehicles
> in the subnet.  If every vehicle on the subnet can expect deliver of
>  multicast packets to every other vehicle on the subnet, then the 
> problem is different than if hidden terminal effects occur or if 
> multicast is unreliable for some other reason.

I think I agree in general.

I would like to discuss details:
- is it more reasonable to put the applicability statement in the IP-
   over-OCB document, or in the vehicular-networking problem statement
   document?  I prefer the  latter.  If you prefer the former, please
   tell where in the document.
- to put the number of vehicles in subnet is reasonable to qualify the
   applicability statement.  This should be in relationship with the max
   power range and the assumed PHY propagation (antennas, etc.)
- the multicast aspects are referred to in the IP-over-OCB draft by
   pointing to [I-D.ietf-mboned-ieee802-mcast-problems.  I think this is
   sufficient.

> Maybe for clarity it would be helpful to include adiagram of a
> general target network with more than three or four vehicles.

There is such a diagram in the vehicular-networking problem statement
draft.  It is Figure 1.  There are successive dots (like '...' expansion
dots)  which suggest there are are more than a few vehicles.

Please read the draft.

Alex

> 
> You can also tell me that my opinion lacks relevance since I have
> not carefully read the draft.
> 
> Regards, Charlie P.
> 
> 
> On 4/15/2019 4:54 AM, Alexandre Petrescu wrote:
>> Hi Brian,
>> 
>> Le 14/04/2019 à 22:49, Brian E Carpenter a écrit :
>>> Hi Alexandre,
>>> 
>>> On 15-Apr-19 04:38, Alexandre Petrescu wrote:
>> 
>> [...]
>>>>> The baseline Neighbor Discovery protocol (ND) [RFC4861] MUST 
>>>>> be used over 802.11-OCB links.  Transmitting ND packets may 
>>>>> prove to have some performance issues.  These issues may be 
>>>>> exacerbated in OCB mode.  Solutions for these problems
>>>>> SHOULD consider the OCB mode of operation.  The best of
>>>>> current knowledge indicates the kinds of issues that may
>>>>> arise with ND in OCB mode; they are described in Appendix J.
>>> 
>>> That's exactly the text that I find problematic. I can't write a 
>>> new version because I lack your expert knowledge, but IMHO it 
>>> should be more specific:
>> 
>> I do have expert knowledge on ND on OCB, although only at a
>> certain level.  It works.
>> 
>> Other people may claim ND does not work on OCB, but have they
>> tried OCB at all?
>> 
>> It is the people who claim ND does not work on OCB should write
>> the new version of the 'TBD' text.
>> 
>> It is very simple to try ND on OCB.  I published a howto on the 
>> email list.  A Hackathon w/o me was tried following that howto.  A 
>> few other organisations that I work with tried it (w/o me).  I did 
>> not received feedback from them about ND not working.
>> 
>> What is difficult to try is to reproduce the ND-over-OCB problem 
>> imagined by Pascal.  It is difficult to try because it involves 4 
>> cars (I dont have that many).  It  is also difficult to try
>> because it seems to miss the fact that OCB can do high power. Maybe
>> that high power solves the problems that appear in ND over low
>> power.
>> 
>> For my part, I cant put text about ND on OCB without having an 
>> implementation of that problem.
>> 
>> It is for that reason that I asked Pascal, or anybody else
>> claiming ND does not work on OCB, to try it and write that text
>> 'TBD'.
>> 
>> Alex
>> 
>>> 
>>> The baseline Neighbor Discovery protocol (ND) [RFC4861] MUST be 
>>> used over 802.11-OCB links.  However, as on any wireless link, 
>>> transmission of multicast ND packets may fail in OCB. In 
>>> particular, scenarios where TBD TBD TBD are likely to be 
>>> unreliable and SHOULD NOT be deployed until an alternative 
>>> standardised solution is available. The best of current
>>> knowledge indicates the kinds of issues that may arise with ND in
>>> OCB mode; they are described in Appendix J.
>>> 
>>> Also I don't like this phrasing in Appendix J:
>>> 
>>> Early experiences indicate that it should be possible to exchange
>>> IPv6 packets over OCB while relying on IPv6 ND alone for DAD and
>>> AR (Address Resolution).
>>> 
>>> Could you rather say the opposite:
>>> 
>>> Early experience indicates that it is possible to exchange IPv6 
>>> packets over OCB while relying on IPv6 ND alone for DAD and AR 
>>> (Address Resolution) in good conditions. However, this does not 
>>> apply if TBD TBD TBD...
>>> 
>>> Regards Brian
>>> 
>>> 
>>>> 
>>>> 
>>>> Alex
>>>> 
>>>>> 
>>>>> Regards Brian
>>>>> 
>>>>> On 14-Apr-19 13:58, NABIL BENAMAR wrote:
>>>>>> +1 Sri
>>>>>> 
>>>>>> On Sun, Apr 14, 2019, 00:06 Sri Gundavelli (sgundave) 
>>>>>> <[e-mail] <mailto:[e-mail]>> wrote:
>>>>>> 
>>>>>> I understand your point Brian, but IMO there are enough 
>>>>>> reasons not to delay this work.
>>>>>> 
>>>>>> There are many use-cases/applications where there is a 
>>>>>> stable topology of RSU¹s and OBU¹s. The regulations around 
>>>>>> 5.9 Ghz (DSRC) band allows the channel use for 
>>>>>> non-priority/non-traffic safety related applications. For 
>>>>>> example, a vehicle in a gas station can receive a coupon 
>>>>>> from the 802.11-OCB radio (AP/RSU) in the gas station. 
>>>>>> There, its a stable topology that classic ND is designed 
>>>>>> for. In this operating mode, its perfectly reasonable to 
>>>>>> use classic ND and it works. The authors have shown enough
>>>>>>  lab data on the same.
>>>>>> 
>>>>>> Ideally, I agree with you that it makes lot more sense to 
>>>>>> publish both the specs at the same time. But, for what
>>>>>> ever reasons the WG went on this path. Authors have spent 
>>>>>> incredible amount of efforts in getting the draft this far 
>>>>>> and we cannot ignore that. You can see the efforts from the
>>>>>> version number; when did we last see a draft version -037?
>>>>>> 
>>>>>> We also need to distill the recent ND discussions and 
>>>>>> filter out the threads that are clearly motivated to
>>>>>> insert a ND protocol that is designed for a totally
>>>>>> different operating environment. An argument that a
>>>>>> protocol designed for low-power environments is the
>>>>>> solution for vehicular environments requires some serious
>>>>>> vetting. Looking at the characteristics, always-sleeping,
>>>>>> occasional internet connectivity, low-power, no memory, no
>>>>>> processing power, no mobility ..etc, meeting vehicular
>>>>>> requirements is some thing most people in the WG do not get
>>>>>> it.
>>>>>> 
>>>>>> Bottom line, IMO, we should move this forward and publish 
>>>>>> the document. All we need is a simple statement in the
>>>>>> spec which puts some scope limits, w.r.t the missing ND
>>>>>> pieces and issues. There are other proposals in the WG that
>>>>>> will address the gaps and bring closure to the work.
>>>>>> 
>>>>>> Sri
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> On 4/12/19, 1:28 PM, "Brian E Carpenter" 
>>>>>> <[e-mail] 
>>>>>> <mailto:[e-mail]>> wrote:
>>>>>> 
>>>>>>> On 13-Apr-19 02:59, Sri Gundavelli (sgundave) wrote:
>>>>>>>> If you go back and check 2017 archives, I did raise 
>>>>>>>> many
>>>>>> of these
>>>>>>>> issues.  But, we clearly decided to limit the scope
>>>>>> excluding address
>>>>>>>> configuration, DAD, ND aspect, link models. When there 
>>>>>>>> is
>>>>>> such a scope
>>>>>>>> statement, it should clearly move these comments to
>>>>>>>> the
>>>>>>>> 
>>>>>> draft that
>>>>>>>> defines how ND works for 802.11-OCB links.
>>>>>>> 
>>>>>>> This is of course possible. In general the IETF hasn't 
>>>>>>> done
>>>>>> that, but has
>>>>>>> followed the lead set by RFC 2464 with the complete
>>>>>> specification of
>>>>>>> IPv6-over-foo in one document.
>>>>>>> 
>>>>>>> However, I don't believe that publishing an RFC about
>>>>>>> the
>>>>>>> 
>>>>>> frame format
>>>>>>> without *simultaneously* publishing an RFC about ND etc
>>>>>> would be a good
>>>>>>> idea. That would leave developers absolutely unable to
>>>>>> write useful
>>>>>>> code, and might easily lead to incompatible
>>>>>> implementations. Since
>>>>>>> we'd presumably like Fords to be able to communicate
>>>>>>> with
>>>>>>> 
>>>>>> Peugeots,
>>>>>>> that seems like a bad idea.
>>>>>>> 
>>>>>>> Regards Brian
>>>>>> 
>>>>> 
>>>>> 
>>>> .
>>>> 
>>> 
>>> 
>> 
>> _______________________________________________ its mailing list 
>> [e-mail] https://www.ietf.org/mailman/listinfo/its