Re: [nvo3] Update on encapsulation design team

[Tom Herbert <[e-mail]>]

On Mon, Oct 24, 2016 at 5:46 PM, Fabio Maino <[e-mail]> wrote:
> Hi Alia,
> please see below.
>
> On 10/24/16 1:44 PM, Alia Atlas wrote:
>
> Fabio,
>
> On Mon, Oct 24, 2016 at 3:54 PM, Fabio Maino <[e-mail]> wrote:
>>
>>
>> I think it will be useful for the design team to articulate clearly and
>> motivate those objections, and present it to the WG for formal discussion
>> BEFORE starting the new design. That should help consolidate the
>> requirements for the design of the new encap.
>>
>> A separate technical document will help  evaluate in depth the objections
>> raised, rather than rely on hundreds of emails and their different
>> interpretation.
>
>
> The WG raised the various objections and I have not seen any efforts to
> claim that specific ones are not accurate technical objections nor that the
> WG should ignore particular objections.
>
>>
>> I think I have seen quite a lot of disagreement around the objection
>> themselves.
>>
>> For example, wrt GPE extensibility the summary reported that
>>
>> "- GPE is insufficiently extensible. Numerous extensions and options have
>> been designed for GUE and Geneve. Note that these have not yet been
>> validated by the WG."
>>
>> That assertion seems to ignore (as it has been pointed out multiple time
>> in the mailing list) what has been done with NSH, or other proposed
>> encodings for carrying metadata with VXLAN-GPE such as:
>> https://tools.ietf.org/html/draft-ietf-sfc-nsh-10
>> https://tools.ietf.org/html/draft-brockners-inband-oam-data
>>
>> I would like the WG to articulate why, from a technical point of view,
>> proper layering can't be used to extend GPE (as recommended by the GPE draft
>> itself), as it is certainly possible (and done) in the two drafts indicated
>> above.
>
>
> It is not the responsibility of the WG to explain why different approaches
> that were not even written down in detail (i.e. an internet-draft) are not
> acceptable solutions.  If you wish to claim that layering with NSH is a
> solution, that may be something interesting to discuss.  However, NSH
> provides for both fixed MD-Type 1 meta-data as well as MD-Type 2 meta-data;
> it assumes a control-plane that communicates what meta-data is to be carried
> and how.    Obviously, an approach that only supports MD-Type 1 meta-data
> will not have the flexibility that I believe was requested for
> extensibility.
>
>
> So, an additional requirement that I infer from your text above is that the
> encap should flexibly allow transport of metadata, even in the absence of a
> control plane. I can't see that requirement spelled out in the summary.
>
> This is one example of what I hope will be part of a detailed analysis of
> what are the requirements of the encap that should be designed. This will
> also help determining what requirements the proposed encaps are not
> matching.
>
>
> I will note that any issues around properly attributing ICMP messages to the
> triggering sender will still exist with large headers.  I do not currently
> understand, from your suggestion, what would be different as far as concerns
> about header size and processing by having extra bytes for the NSH header.
>
>
> I think we are all quite familiar with the alleged simplicity of the
> hand-waved design option.
>
>
> Right, that's why I'm suggesting to avoid to hand-wave requirements.
>
>
>
>>
>> Same goes for the security objection to GPE.
>
>
> Frankly, I can't parse this other than assuming you mean that security is
> not required.  Again, if you do not think the WG should agree with this
> technical objection, then you need to be able to clearly explain why and see
> if there is WG consensus.
>
The need for extensibility came out of the requirements for nvo3
security which are listed in draft-ietf-nvo3-security-requirements.
We already had a discussion on the security objections raised for
VXLAN-GPE in the "Security in VXLAN-GPE" thread (security concerns
about VXLAN have also been raised elsewhere also
http://blog.ipspace.net/2015/04/omg-vxlan-encapsulation-has-no-security.html).
I still don't see how data plane security requirements can be
satisfied in VLXAN-GPE-- again the "just use NSH" argument provides
nothing substantive for a protocol specification. Neither do I believe
this is something we can plan to address down the road, security is a
real concern now for those of us deploying protocols at scale.

I was assuming that the design team will carefully consider how to
best meet the requirements around security and the closely related
issues around DOS attacks (hopefully last week's attacks are a wakeup
call to this), but maybe this should be explicit in the DT charter to
avoid confusion. IMO getting security right is far more important than
the goal of trying to maintain compatibility with existing protocols.

Tom

>
>>
>> I'm quite sure the authors of the other two proposals have similar view on
>> the objections raised to their proposals.
>
>
> Then they - and others - are and have been welcome to clearly express on the
> NVO3 mailing list why those objections should be ignored.   That is an
> option - if there is WG consensus on it.  PLEASE reread RFC 7282.
>
>>
>> A separate document, in my opinion, would help understanding what needs to
>> be addressed before we start with the new design.
>
>
> Do you want to have an IETF standard in this space?
>
>
> It doesn't have to become a standard, but I think it should be done, and
> discussed in the WG, before the design begins. Having it as a separate draft
> might help. If then you want to merge it or morph it into the design doc
> that will be fine.
>
>
> I understand the desire to stall progress in favor of  proprietary
> implementations.
>
>
> If you are referring to my desires, I'm afraid you are grossly
> misunderstanding.
>
> I would prefer that the IETF not fail to provide standards in this space.
>
>
> Right, we just happen to disagree on how those standards would look like,
> and how we can get there.
>
> FWIW, I'm  just answering to the call of the chairs to "hear opinions and
> confirmation or disagreement on interest in creating a DP encapsulation that
> addresses the various technical concerns".
>
> Thanks,
> Fabio
>
>
>
> I am not willing to leave open a WG that is determinedly failing to make
> progress on its charter.
>
>
> Regards,
> Alia
>
>
>>
>>
>> Thanks,
>> Fabio
>>
>>
>> On 10/24/16 12:51 PM, Anoop Ghanwani wrote:
>>
>> Hi Alia,
>>
>> Ideally I would rather not go there at all as expressed in the first part
>> of Fabio's note.  Assuming we must go there, having a section that documents
>> why the WG thought we needed a new one would be good enough for me.
>>
>> Anoop
>>
>> On Mon, Oct 24, 2016 at 11:46 AM, Alia Atlas <[e-mail]> wrote:
>>>
>>> Hi Anoop,
>>>
>>> On Mon, Oct 24, 2016 at 2:43 PM, Anoop Ghanwani <[e-mail]>
>>> wrote:
>>>>
>>>> Alia,
>>>>
>>>> I think it will provide an official reference which will be helpful
>>>> since all the encaps will be around a long time and we can point people to
>>>> that document when we are asked the question "why did they develop yet
>>>> another encap?".
>>>
>>>
>>> So what you are asking for is a section in the overall draft that talks
>>> about the motivations and improvements?
>>> There's a trade-off of speed and getting a solution done versus doing
>>> process work for a theoretical future that won't happen if we don't get the
>>> technical work finished.
>>>
>>> Regards,
>>> Alia
>>>
>>>
>>>>
>>>> Thanks,
>>>> Anoop
>>>>
>>>> On Mon, Oct 24, 2016 at 11:30 AM, Alia Atlas <[e-mail]> wrote:
>>>>>
>>>>> Anoop & Fabiio,
>>>>>
>>>>> On Mon, Oct 24, 2016 at 2:26 PM, Anoop Ghanwani <[e-mail]>
>>>>> wrote:
>>>>>>
>>>>>> Agree with Fabio (including the suggestion for an interim deliverable
>>>>>> on shortcomings).  If the WG doesn't agree on the shortcomings, chances are
>>>>>> they may not like the 4th encap.
>>>>>
>>>>>
>>>>> What do you expect to be different from the summary of technical
>>>>> objections that came out of the last discussion?  Are you looking for more
>>>>> detail?
>>>>>
>>>>> I didn't see disagreement about the accuracy of the technical
>>>>> objections.
>>>>>
>>>>> Regards,
>>>>> Alia
>>>>>
>>>>>>
>>>>>> Anoop
>>>>>>
>>>>>> On Thu, Oct 20, 2016 at 12:10 PM, Dino Farinacci <[e-mail]>
>>>>>> wrote:
>>>>>>>
>>>>>>> I agree with Fabio.
>>>>>>>
>>>>>>> Choosing a single encapsulation that is not 1 of the 3, creates a 4th
>>>>>>> one that no one wants.
>>>>>>>
>>>>>>> And guess what, you make all 3 authors unhappy where none of them
>>>>>>> will endorse (or implement) the 4th one.
>>>>>>>
>>>>>>> Dino
>>>>>>>
>>>>>>> > On Oct 20, 2016, at 12:02 PM, Fabio Maino <[e-mail]> wrote:
>>>>>>> >
>>>>>>> > (for full disclosure I'm one of the authors of VXLAN-GPE)
>>>>>>> >
>>>>>>> > Matt, Sam, Alia,
>>>>>>> > I've expressed multiple times and in multiple venues my adversity
>>>>>>> > (and the motivations) to set this group to design yet another encapsulation.
>>>>>>> > I won't repeat it here once again, but I want to re-assert that it's still
>>>>>>> > were I stand.
>>>>>>> >
>>>>>>> > I've seen quite a few people in the mailing list here expressing
>>>>>>> > similar concerns, but I see that it has not changed the opinion of the
>>>>>>> > chairs and the AD on what they believe is the best way to move forward.
>>>>>>> >
>>>>>>> > That said, here are my comments to the charter.
>>>>>>> >
>>>>>>> > I think the design team first goal should be to clearly articulate
>>>>>>> > the shortcomings of the current encapsulations proposed to the WG. This
>>>>>>> > should be the very first deliverable of the design team. The actual design
>>>>>>> > work should start only once the WG has reached consensus on that document.
>>>>>>> > Especially considering that some of the encapsulations proposed are being
>>>>>>> > deployed, I think articulating the shortcomings will help to make the best
>>>>>>> > choice in term of (1) selecting which one will need to be extended, and (2)
>>>>>>> > designing the actual extensions.
>>>>>>> >
>>>>>>> > Below are my proposals on how to modify the wording of the charter.
>>>>>>> >
>>>>>>> >
>>>>>>> >
>>>>>>> > On 10/20/16 1:37 AM, Bocci, Matthew (Nokia - GB) wrote:
>>>>>>> >> WG,
>>>>>>> >>
>>>>>>> >> We would like to give you an update on the process in the WG for
>>>>>>> >> progressing the issue of a data plane encapsulation. The chairs and Alia
>>>>>>> >> believe that the best way forward is to progress a single encapsulation
>>>>>>> >> format that addresses the technical concerns raised on the list in the
>>>>>>> >> recent discussions. This would address the clear overall consensus of the
>>>>>>> >> Berlin meeting and list for a single encapsulation.
>>>>>>> >>
>>>>>>> >> The strategy should be to take one of the three existing
>>>>>>> >> encapsulations and enhance it to address these concerns. This would become
>>>>>>> >> the standards track output of the WG. The existing three drafts (GENEVE, GUE
>>>>>>> >> and VXLAN-GPE) should be forwarded to the IESG as informational after the
>>>>>>> >> standards track draft specifying the single encapsulation. This provides an
>>>>>>> >> opportunity for those encapsulations to be documented and maintained.
>>>>>>> >>
>>>>>>> >> The single encapsulation should be viewed as one that the WG and
>>>>>>> >> industry can converge around for the future.
>>>>>>> >>
>>>>>>> >> We have created a design team to progress work on a single
>>>>>>> >> encapsulation that can form the basis or work going forward. The design team
>>>>>>> >> members are: Michael Schmidt, Uri Elzur, Ilango Ganga, Erik Nordmark, Rajeev
>>>>>>> >> Manur, Prankaj Garg. Many thanks to these individuals for their help.
>>>>>>> >>
>>>>>>> >> Please see below for a draft charter for the design team. Please
>>>>>>> >> review the charter and send comments to the list by 2nd November 2016.
>>>>>>> >>
>>>>>>> >> Regards,
>>>>>>> >>
>>>>>>> >> Matthew and Sam
>>>>>>> >>
>>>>>>> >>
>>>>>>> >> ====
>>>>>>> >> NVO3 Encapsulation Design team 2016
>>>>>>> >>
>>>>>>> >> Problem Statement
>>>>>>> >> The NVO3 WG charter states that it may produce requirements for
>>>>>>> >> network virtualization data planes based on encapsulation of virtual network
>>>>>>> >> traffic over an IP-based underlay data plane. Such requirements should
>>>>>>> >> consider OAM and security. Based on these requirements the WG will select,
>>>>>>> >> extend, and/or develop one or more data plane encapsulation format(s).
>>>>>>> >>
>>>>>>> >> This has led to drafts describing three encapsulations being
>>>>>>> >> adopted by the working group:
>>>>>>> >> - draft-ietf-nvo3-geneve-03
>>>>>>> >> - draft-ietf-nvo3-gue-04
>>>>>>> >> - draft-ietf-nvo3-vxlan-gpe-02
>>>>>>> >>
>>>>>>> >> Discussion on the list and in face-to-face meetings has identified
>>>>>>> >> a number of technical problems with each of these encapsulations.
>>>>>>> >> Furthermore, there was clear consensus at the IETF meeting in Berlin that it
>>>>>>> >> is undesirable for the working group to progress more than one data plane
>>>>>>> >> encapsulation. Although consensus could not be reached on the list, the
>>>>>>> >> overall consensus was for a single encapsulation (RFC2418, Section 3.3).
>>>>>>> >> Nonetheless there has been resistance to converging on a single
>>>>>>> >> encapsulation format, although doing so would provide the best benefit to
>>>>>>> >> the industry.
>>>>>>> >
>>>>>>> > The portion of the last sentence that follows the comma ("although
>>>>>>> > doing so would provide the best benefit to the industry") doesn't seem to be
>>>>>>> > adding anything to the charter. I'd suggest it could be removed.
>>>>>>> >
>>>>>>> >>
>>>>>>> >> Design Team Goals
>>>>>>> > The design team should clearly articulate in a draft which are the
>>>>>>> > shortcomings of the proposed encapsulations, and where they fall short in
>>>>>>> > addressing the NVO3 architectural requirements.
>>>>>>> >
>>>>>>> > Once the 'shortcomings' draft has reached consensus of the WG,
>>>>>>> >> The design team should take one of the proposed encapsulations and
>>>>>>> >> enhance it to address the technical concerns.
>>>>>>> >> Backwards compatibility with the chosen encapsulation and the
>>>>>>> >> simple evolution of deployed networks as well as applicability to all
>>>>>>> >> locations in the NVO3 architecture
>>>>>>> > , together with the design goals articulated in the 'shortcoming'
>>>>>>> > draft,
>>>>>>> >
>>>>>>> >> are goals. The DT should specifically avoid a design that is
>>>>>>> >> burdensome on hardware implementations, but should allow future
>>>>>>> >> extensibility. The chosen design should also operate well with ICMP and in
>>>>>>> >> ECMP environments. If further extensibility is required, then it should be
>>>>>>> >> done in such a manner that it does not require the consent of an entity
>>>>>>> >> outside of the IETF.
>>>>>>> >>
>>>>>>> >> Timeline
>>>>>>> >> The design team should
>>>>>>> > first produce the 'shortcomings' draft, get it adopted by the WG,
>>>>>>> > and then
>>>>>>> >
>>>>>>> >> produce a first draft describing the proposal by end of January
>>>>>>> >> 2017. Target adoption by the WG by March 2017 IETF.
>>>>>>> >>
>>>>>>> > (those two dates may need to be adjusted accordingly)
>>>>>>> >
>>>>>>> >
>>>>>>> > Thanks,
>>>>>>> > Fabio
>>>>>>> >
>>>>>>> >>
>>>>>>> >>
>>>>>>> >>
>>>>>>> >>
>>>>>>> >>
>>>>>>> >>
>>>>>>> >> _______________________________________________
>>>>>>> >> nvo3 mailing list
>>>>>>> >>
>>>>>>> >> [e-mail]
>>>>>>> >> https://www.ietf.org/mailman/listinfo/nvo3
>>>>>>> >
>>>>>>> >
>>>>>>> > _______________________________________________
>>>>>>> > nvo3 mailing list
>>>>>>> > [e-mail]
>>>>>>> > https://www.ietf.org/mailman/listinfo/nvo3
>>>>>>>
>>>>>>> _______________________________________________
>>>>>>> nvo3 mailing list
>>>>>>> [e-mail]
>>>>>>> https://www.ietf.org/mailman/listinfo/nvo3
>>>>>>
>>>>>>
>>>>>>
>>>>>> _______________________________________________
>>>>>> nvo3 mailing list
>>>>>> [e-mail]
>>>>>> https://www.ietf.org/mailman/listinfo/nvo3
>>>>>>
>>>>>
>>>>
>>>
>>
>>
>
>
>
> _______________________________________________
> nvo3 mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/nvo3
>