Re: [Isis-wg] [Bier] BAR field length in draft-ietf-bier-isis-extensions and draft-ietf-bier-ospf-extensions

[IJsbrand Wijnands <[e-mail]>]

Alia,

> I am quite disappointed in the level of discourse; the BIER-WG has traditionally worked very
> collegially with substantive technical points.

I don’t think its necessary to try and discredit people like this for speaking up.

If you ask for input from the WG, you should be honest enough to list all the options.

What Les and I say is because we think that is architecturally the right think to do. If you don’t understand the reasoning or disagree, that is fine. 

Some problems take time to converge on, clearly in this case there was not enough time, but the drafts are pushed through anyway, your call. We got informed through a draft that there was no consensus on how to use the BAR type 4 weeks ago.

Les’s response to Andrew’s comments are spot on. Over the weekend we tried to converge over a 16 bit BAR, and how it would look like. We where not able to converge on the semantics, especially related to option B and the “BAR sub-type”. Its opening an other can of worms and will cause new discussion over the BIER architecture in the future. I know, not your problem.

Thx,

Ice.

> 
> Regards,
> Alia
> 
> On Mon, Feb 19, 2018 at 10:11 PM, Alia Atlas <[e-mail]> wrote:
> Les,
> 
> On Mon, Feb 19, 2018 at 9:32 PM, Les Ginsberg (ginsberg) <[e-mail]> wrote:
> I am very sympathetic to doing “as little as possible” given we are talking about documents which are going through final reviews.
> 
> At the same time, I think defining the authoritative source for algorithm values is important.
> 
>  
> 
> I therefore agree w Ice – let’s keep the current 8 bit algorithm value – but make it clear that the identifiers come from the IGP Registry. 
> 
> 
> I do not hear from you or Ice a clear technical reason nor willingness to address the concerns that I raised about the impact
> on the use of BIER. 
> 
> I see no technical reasons being used to recommend combining the BIER BAR registry and the IGP Algorithms registry.
> 
> 
> Andrew – I do not think there is agreement on what the function of a “BAR sub-type” is. Therefore I am not comfortable in adding it to the drafts.
> 
> Certainly this may prove to be useful, but let’s add it when we know how it will be used and how to assign values to it. That requires more discussion than can reasonably be had in the current context.
> 
>  
> 
> Tony / Alia – the argument that 256 algorithm values is not enough for all use cases (BIER specific and IGP specific and Babel specific…) – or even that 128 is not enough (if we allow the Flex-Algo proposal to reserve half of the space) – simply does not ring true to me.
> 
> If I waited for  you to buy  me a beer when we reached 10 algorithms I likely will go thirsty for a very long time. 
> 
> 
> It is fascinating to see that you believe me too busy to keep up on the side discussions happening - or perhaps merely too distracted to
> recall the emails discussing acquiring half of the IGP Algorithm space for flex-algo.  I do talk to my WG Chairs.
> 
> Please engage substantively with technical arguments.  If you have them, you are more than capable of representing them
> accurately.
> 
> 
> So Option E seems best to me.
> 
> 
> That was not part of my listed options.
>  
> Regards,
> Alia
> 
> 
>  
> 
>    Les
> 
>  
> 
> From: BIER [mailto:[e-mail]] On Behalf Of Tony Przygienda
> Sent: Monday, February 19, 2018 6:20 PM
> To: Dolganow, Andrew (Nokia - SG/Singapore) <[e-mail]>
> Cc: BIER WG <[e-mail]>; IJsbrand Wijnands <[e-mail]>; [e-mail] list <[e-mail]>
> Subject: Re: [Bier] [Isis-wg] BAR field length in draft-ietf-bier-isis-extensions and draft-ietf-bier-ospf-extensions
> 
>  
> 
> My reaction to AD's options:
> 
> IF you prefer or can accept the current status, but think there should be an IANA registry
> as is usual for managing code-points, please say so.  No more justification is needed.
>  
> 
>  
> 
> +1 to this option, i.e. current status with IANA BIER BAR registry. 
>  
> I think we have a clear and current case which is anchored already in the architecture RFC as section 6.9
>  
> The draft https://tools.ietf.org/html/draft-zzhang-bier-algorithm-00 goes into more detail in this respect and I hope the draft being adopted and at least a single BAR value being assigned to deal with brownfield deployments today where not all routers support BIER.  This clearly necessitates a BIER BAR registry. 
>  
> 
> IF you prefer Option B, C, and/or D, please say so with your explanation.  More technical depth than "'we might need it" would be helpful; the availability of sub-TLVs already
> provides future proofing.
> 
>  
> +2 for Option B). Albeit we can meet future use cases with BAR subtype being a sub-TLV a bar type/bar subtype 16 bits field (subtype with or without registry and of course we can name those things differently) all sub-TLVs in IGPs are traditionally “optional”, i.e. adding sub-TLVs later _may_ cause backwards compatibility problems.  Defining the subtype today will allow us to specify that only type/subtype 0/0 is well defined & any unknown type/subtype combination must be avoided. The subtype will allow for BAR 0 or future BAR types to understand that subtype is in a sense a “mandatory sub-TLV” and the routers sending out such subtype must be avoided, even if the type itself is known.  Several possibilities come to mind immediately such as using Unicast IGP Algorithm Registry as subtype for certain BAR values or accommodate interesting, future work like Flex Algo into BIER right after IGP RFCs are available which IMO would present a very beneficial future direction for BIER technology cleanly governed by a BIER BAR registry.
> 
> PS: I think we should control the urges of workgroup participants to explore the number of letters in the roman alphabet that we can use to introduce their preferred solutions if we want to get to some kind of consensus on this thread.
> 
> PPS: Option E has been discussed in the last 16-bit thread to the point of logical conclusion of a "all routing protocols under the sun algorithm registry" and found to  be a rathole I thought ...
> 
>  
>  
> 
>  
> 
> On Mon, Feb 19, 2018 at 6:09 PM, Dolganow, Andrew (Nokia - SG/Singapore) <[e-mail]> wrote:
> 
> All,
> 
>  
> 
> As we discussed here (as a WG) and in this topic:
> 
> 	• We need to have ability to define way for independent BIER computation algorithms (for BIER specific computations or other use cases, some of which Alia highlighted in her email below)
> 	• We want to have extensibility to use other non-BIER specific algorithms (as others brought up)
>  
> 
> The original draft can be argued not to provide both of those capabilities, and thus Option A below (marked as Current status) really just defers the issue. I find Option E that Ice added also counterproductive as it eliminates the top use case above. Thus we really left with options B, C, D as a compromise. From those, Option B seems to me the best:
> 
> 	• It meets the requirements above
> 	• It allows a clean implementation (as opposed to Option C which is a bit more kludgy). Thanks to a sub-TLV defining what BAR field carries – a BIER-specific algorithm defined in BIER specific registry (a registry that should be BIER specific, regardless of IGP used), or something else to meet needs expressed by others, we meet the requirements from those who wanted to change the Current status
>  
> 
> Option C/D are acceptable alternatives; however, Option B seems technically cleanest, most flexible, and meeting all requirements.
> 
>  
> 
> Andrew
> 
>  
> 
>  
> 
> From: Isis-wg <[e-mail]> on behalf of Alia Atlas <[e-mail]>
> Date: Tuesday, February 20, 2018 at 12:05 PM
> To: "(Ice) IJsbrand Wijnands" <[e-mail]>
> Cc: BIER WG <[e-mail]>, "[e-mail] list" <[e-mail]>
> Subject: Re: [Isis-wg] [Bier] BAR field length in draft-ietf-bier-isis-extensions and draft-ietf-bier-ospf-extensions
> 
>  
> 
> Ice,
> 
>  
> 
> On Mon, Feb 19, 2018 at 7:57 PM, IJsbrand Wijnands <[e-mail]> wrote:
> 
> Alia,
> 
>  
> 
> I appreciate that you have finally decided to discuss this on the BIER mailing list.
> 
> 
> I know that there are individual drafts draft-ppsenak-ospf-sr-flex-algo-00  and  draft-hegdeppsenak-isis-sr-flex-algo-02.
> I see a bit of discussion on the is-is mailing list and at IETF 100, but, of course, no WG adoption.
> 
> I see BIER as a fundamental technology that can be used in different situations.  For instance, there is not merely 
> discussion of how Babel and BIER could interact - but actual code (thanks Sandy!); of course, that is not a WG-adopted
> draft yet either, so this is merely a thought experiment example.  How do the different algorithms
> work for an IGP that isn't link-state?   What about the ideas around using BIER with caches?  Are there issues there? 
> What about algorithms that make sense for BIER or multicast - but not for unicast?
> 
> IANA registries are not price prohibitive.  Why would we tie BIER to the link-state IGP registry?
> 
>  
> 
> We are talking about what needs to be advertised in OSPF and ISIS in order to select the BIER underlay. We are not discussing Babel or any other candidate underlay technologies for BIER. Moreover, we are not limiting any new innovation with BIER regarding the underlay. This discussion is strictly related to the drafts in the title.
> 
>  
> 
> I do not hear you making a technical argument.
> 
>  
> 
> This is an architectural argument!
> 
>  
> 
> An architectural argument can't also limit itself to the drafts in the title.
> 
>  
> 
> If it sounded like the IANA registry was suggested as separate for BIER OSPF  and BIER ISIS, then your attempt to reframe the conversation might be reasonable.  Let me clarify - I see no current reason for an OSPF BAR registry and an ISIS BAR registry; it would just be a BAR registry.  Perhaps
> 
> that clarification is a good reason to get the IANA registry included in the next update?
> 
>  
> 
> The routing layer is separate from the BIER layer.  The BAR is for the BIER layer.
> 
>  
> 
> Regards,
> 
> Alia
> 
>  
> 
>  
> 
> Hope this clarifies,
> 
>  
> 
> Thx,
> 
>  
> 
> Ice.
> 
>  
> 
> 
> Regards,
> Alia
> 
> 
> On Mon, Feb 19, 2018 at 7:03 PM, IJsbrand Wijnands <[e-mail]> wrote:
> Hi Alia,
> 
> There is one more option that I think is not fully covered from the choice of options related to getting a registry.
> 
> The topic of the discussion is what information we need to pass in the IGP in order for BIER to select the correct underlay. What identifies the underlay is really what ever information is needed to select the Table (MT-ID) and Algorithm. An example of Algorithm work that is going on is Flex-Algo. My preferred option is to align with what ever the IGPs are using to identify the Algorithm.
> 
> Option E: Change BAR into “IGP Algorithm” registry as documented in https://www.iana.org/assignments/igp-parameters/igp-parameters.xhtml#igp-algorithm-types
> 
> Thx,
> 
> Ice.
> 
> 
> On 19 Feb 2018, at 13:51, Alia Atlas <[e-mail]> wrote:
> 
> As the Sponsoring AD for draft-ietf-bier-isis-extensions-07 and draft-ietf-bier-ospf-extensions-12, I have been following the discussion on the mailing list with interest.
> 
> I have not seen clear consensus for any change.
> 
> Let me be clear on what I see the options are from the discussion.  Then I'll elaborate
> a bit on how you can express your perspective most usefully.
> 
> 1) Current Status:  Bier Algorithm (BAR) field is 8 bits.  Currently, only value 0 is specified.  The drafts do not have an IANA registry - with the expectation that one will be created when the first additional use is clear.  It is possible that there will be objections from the IESG to progressing without an IANA registry.  Given the lack of clarity for future use-cases and after discussion, I decided not to force one after my AD review - but I will not push back against having a BIER IANA registry if raised by others.
> 
> 2) Option B:  Add a BAR sub-type of 8 bits.  This would modify the current TLVs.
>    Define an IANA registry for the BAR type.  The meaning of the BAR sub-type derives 
>    from the BAR type.   We can debate over the registration policy for the BAR type.
> 
> 3) Option C: Change the BAR field to be 16 bits and define an IANA registry.  Part of the range can be FCFS with Expert Review, part can be Specification Required, and part can be IETF Consensus.
> 
> 4) Option D: At some point in the future, if there is an actual understood and documented need, a BAR sub-type could be added a sub-TLV.  The length of the BAR sub-type could be determined when the sub-TLV is defined.
> 
> Given
> 
>   a) option D exists 
>   b) there is currently only one defined value for BAR
>   c) I do not see strong consensus for change to one particular other option
> 
> I see no current reason for a change and I certainly see absolutely no reason for
> a delay in progressing the documents.
> 
> I do want to be clear about what the WG wants to do on this issue.  Therefore, here is
> my following request.
> 
> Please send your feedback to the mailing list as follows:
> 
> IF you prefer or can accept the current status, please say so.  No more justification
> or reasoning is required. I just don't want the bulk of folks who are content to be
> overlooked by those suggesting change.
> 
> IF you prefer or can accept the current status, but think there should be an IANA registry
> as is usual for managing code-points, please say so.  No more justification is needed.
> 
> IF you prefer Option B, C, and/or D, please say so with your explanation.  More technical depth than "'we might need it" would be helpful; the availability of sub-TLVs already
> provides future proofing.
> 
> IF you have a clear technical objection to why the Current Status is not acceptable,
> please express that - with clear details.
> 
> IF you feel that additional code-points should be allocated in a BAR IANA Registry or
> have thoughts on the appropriate policy, please say so with your explanation for what
> those should be.
> 
> Unless I see clear and strong consensus for something other than the Current Status,
> that will remain.
> 
> IF there is clear and strong consensus for Option B, C, or D, or adding an IANA registry with particular values, then it will be possible to have a change up through this Weds night - with a 1 week WGLC on that particular technical change.
> 
> My priority is to have the base BIER specifications published as Proposed Standards so that more BIER implementations and deployment can be done.  I would like the WG to wrap up the core work (as expressed in the proposed recharter) so that you all can look
> at how to use it.
> 
> Given this topic was raised last Weds and given that there are no technical objections raised to the documents as are, there isn't much time - so please just respond to this email ASAP.  My deadline for a decision is 6pm EST on Weds.
> 
> Regards,
> Alia
> 
> _______________________________________________
> BIER mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/bier
> 
>  
> 
> <PastedGraphic-6.png>
> 
>  
> 
> <image001.png>
> 
>  
> 
>  
> 
> 
> _______________________________________________
> BIER mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/bier
> 
>  
> 
> 
> _______________________________________________
> Isis-wg mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/isis-wg
> 
> 
>