Re: [Roll] Way forward for draft-clausen-lln-rpl-experiences

[C Chauvenet <[e-mail]>]

Thomas,
See inline,  

Le 16 mai 2012 à 17:08, Thomas Heide Clausen a écrit :

> Dear Cedric,
> 
> Thank you for your email. Comments below.
> 
> On 16 May 2012, at 16:13, C Chauvenet <[e-mail]> wrote:
> 
>> Hi, 
>> 
>> I definitely agree that implementation feedback is always good to know, so your experiences are welcomed.
>> 
>> I also think that problems investigations need a complete and exact view, so I would encourage you to put much more details about the scenario and the environment where you experimentations took place.
>> For instance, I would enjoy a "RPL Implementation Description" section in you draft listing the hardware your used, your RPL parameters, the RPL drafts and mechanisms implemented, your OS etc...
> 
> As I indicated to JP already:
> 
> 1) 	this draft is not "just" based on a single scenario, environment or implementation 
> 	(and, therefore, not "just" based on a single test). 

Does it means that your run multiple scenarios ? 
That would be great.

> 
> 2) 	the observations that we made during the test_S_ served to make us look at the RPL
> 	RFC again, to explain our observations and to generalize them from the written word
> 	of that RFC.
> 
> Therefore, I do not think that any such description would bring anything (other than entropy and delay) to the process. The RPL RFC and this I-D is more than sufficient.
> 
> As to what features were implemented, that is easy to answer: the RPL RFC (except for security) to the letter and with the parameters suggested there, and OF0. However, even the parameters are immaterial for the observations listed in this I-D.

I think we are progressing on the definition of your RPL implementation : could you also precise what part of the RPL spec you have implemented ? eg. What mode of Operation and why, what options did you choose to include in DIO messages, your metrics ....

> 
>> If I read a paper with orthogonal observations with the same level of details as in your draft, then how could I forge my opinion ?
> 
> I'd suggest reading the indicated parts of the RPL RFC conjointly with this I-D. Again, the observations that we present in this I-D make exclusive reference to RPL, and not to a specific implementation or deployment.
> 
>> Looking at this draft, it seems that it gathers lots of previous discussions that occurred during the past year on various mailing lists, and IETF meetings.
>> 
>> Does your experimentations takes care about these recommendations ?
> 
> Again, the issues raised in this I-D are based on what can be observed from the RPL RFC. 
> 
> If there are additional considerations required (which you seem to indicate to be the case) which are not reflected in the RPL RFC, in order to overcome the issues raised, then that indeed should be a big problem for the IETF and for ROLL....

These recommendations were for the problems you pointed out in *Your* implementation.
Of course if other implementers are facing to the same problems, they could rely on it, but I have not heard about similar problems outside your lab from now.

> 
>> If not, does your draft mention the propositions that have been made to address the problems you point out in your draft ?
>> I think it could be worth to leverage on these previous discussions.
> 
> I firmly disagree. The IETF and ROLL has published an RFC - that's what this I-D discusses. 
> 
> Discussing what may have been proposed in other media would be entirely out of scope.

Again, this is not related to the RPL RFC but *your* implementation that received some guidance regarding its problems.
The RPL RFC should not include tips for your implementation, but your implementation and your draft should pay attention to the tips given to you.

> 
>> Your draft is a list of Description and Observations.
>> Maybe you could add a "Resolution Proposal" section for each problem, gathering previous discussion and your own proposals ?
> 
> Nope. That's not the objective of this I-D.

That's too bad, this is what all readers of this draft are looking for !
Without offense, saying "this is bad" without a better proposition seems like half-work ! 
And this is true for most of the discussion topics in life, far away from RPL !

> I would venture that if the WG is serious about applicability statements, then those applicability statements would be the place for discussion of "how this issue, raised in this I-D, is addressed or moot for a given deployment".
> 
>> Identifying what is wrong in your implementation
> 
> Considering that these observations are not implementation-specific, but are directly on the RPL RFC, I venture the observations that there's nothing "wrong in my implementation" here.
> 

As a general comment : RPL is a routing protocol targeted for a very wide application area. 
Some may think this is good because it covers a lot of needs.
Some may think it is bad because it is wide and not specific enough for their application.
Wathever your position, these two arguments are valid, this is all a matter of viewpoint and tradeoff.
So, because RPL is wide enough for multiple application, you have to take some time to tune it correctly, according to your application.
One simple choice that every RPL implementers will ave to answer is : Use Storing or Non Storing Mode ? This is a fondamental design choice that cannot be made outside a scenario consideration.
Be sure that there is no magic RPL tuning that works for all, but multiple fine RPL tuning for multiple applications.
I honestly cannot believe in generic results given the incredibly variety of tuning that RPL can have.
So my final position is that we disagree on the need of more details in your experiments.

Cédric.

> Thomas
> 
> 
>> is a good first step, but the hardest part is to propose some corrections.
>> Best Regards,
>> 
>> Cédric Chauvenet.
>> 
>> Le 16 mai 2012 à 15:04, JP Vasseur a écrit :
>> 
>>> Dear Thomas,
>>> 
>>> On May 16, 2012, at 2:08 PM, Thomas Heide Clausen wrote:
>>> 
>>>> Dear JP and Michael,
>>>> 
>>>> Thank you for your mail.
>>>> 
>>>> On May 16, 2012, at 09:18 , JP Vasseur wrote:
>>>> 
>>>>> Dear Thomas,
>>>>> 
>>>>> On May 11, 2012, at 8:25 AM, Thomas Heide Clausen wrote:
>>>>> 
>>>>>> Dear JP, Michael, all
>>>>>> 
>>>>>> Upon JPs invitation, draft-clausen-lln-rpl-experiences was presented and discussed at the Paris meeting.
>>>>>> 
>>>>>> The authors consider the document complete and "done", and are looking to take it forward in the IETF 
>>>>>> process for publication as "Informational RFC" in the very near future. 
>>>>>> 
>>>>>> We would therefore like to ask the WG chairs, if the ROLL WG is willing to accept and progress this 
>>>>>> document towards publication?
>>>>> 
>>>>> Thanks for your suggestion. So far we haven't see a lot of discussion/interest from the WG but your request is
>>>>> perfectly fair.
>>>> 
>>>> Thank you - I aim to be fair.
>>>> 
>>>>> So far there are no details on the scenarios and testing environments that led to the issues that 
>>>>> you reported, thus I would suggest you to first include them so that people interested could be able to reproduce
>>>>> it. Once the drat is updated, we'll be happy to pool the WG.
>>>>> 
>>>>> Does that make sense ?
>>>> 
>>>> Not really. Let me explain my disagreement.
>>>> 
>>>> We tried RPL (and, I note, several different independent implementations of RPL) in a number of different scenarios and deployments, and observed the behaviors exhibited - noting that what we observed across the different implementations, scenarios and deployments was fairly universal.
>>>> 
>>>> We then went back to the specification, to understand these behaviors in detail, and understand their universality as independent from a specific scenario or deployment or implementation - but rather, as artifacts of the RPL protocol design.
>>>> 
>>>> We therefore believe that _any_ deployment, scenario or testing environment of RPL needs to pay attention to the issues presented, and find a way of addressing them. The way of addressing these issues in a given deployment or scenario would be appropriate for an "applicability statement" for that deployment or scenario.
>>> 
>>> JP> Thanks for the clarification; that being said, for the WG to make sure that nothing is "scenario" dependent and the outcome could indeed apply to all scenarios,
>>> it might be worth being more explicit. For example, you pointed out to the MTU issue, to which I mentioned that 15.4g would bring a solution, so you may want to 
>>> explain that you did not use 15.4g and there are a number of such examples ….
>>> 
>>>> 
>>>> (For example, a deployment using only L2s which provides guaranteed bi-directional links  for L3 would address this by in the applicability statement stating "As all L2-links are guaranteed bi-directional, this addresses the issues raised in section 9 in draft-clausen-lln-rpl-experiences".)
>>>> 
>>>> Thus, we believe that it would actually be misleading (not to mention, unnecessarily verbose) to put the "details on the scenarios and testing environments" into this I-D.
>>>> 
>>>> Doing so would mislead the reader to believe that the issues presented only manifest themselves in those precise scenarios - which definitely isn't the case.
>>> 
>>> JP> see the previous comment and tell us what you think; we could provide other examples.
>>> 
>>> Note that we do not oppose to asking to the WG; we just request you first to add additional information to proceed forward.
>>> 
>>> thanks.
>>> 
>>> JP and Michael.
>>> 
>>> JP.
>>> 
>>>> 
>>>> Best,
>>>> 
>>>> Thomas
>>>> 
>>>>> Thanks.
>>>>> 
>>>>> JP and Michael.
>>>>> 
>>>>>> 
>>>>>> Best,
>>>>>> 
>>>>>> Thomas, Ulrich, Yuichi, Jiazi and Axel
>>>>> 
>>>> 
>>> 
>>> _______________________________________________
>>> Roll mailing list
>>> [e-mail]
>>> https://www.ietf.org/mailman/listinfo/roll
>>> 
>> 
>> 
>