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

[Tony Przygienda <[e-mail]>]

First, I concur with Alia's writeup fully and yes, this is definitely
providing a "higher ground" and clean architecture what those 16 bits are
aiming to achieve if they really should serve their purpose in the oncoming
years. Beside providing a nice orthogonal framework to think about those
issues it gives BIER the possibility to use the underlying IGP as any
combination of (signalling, computation algorithm, constraints)  so to
speak while we can substitute/add the algorithm and/or constraint part with
BIER information/algorithm as needed in each use case. Efficient
implementation is a different topic, at minimum we understand it for things
like SPF quite well from other technologies. The only thing BIER will never
try to do is topology distribution when using IGP of course  ;-)   If we
took all this fire to the forge here to emerge with that framework as
consensus, it may have been very well worth it IMO.

Second, I concur with Jeffrey terminology-wise that the BART is the BIER
algorithm/constraint side normally when we talk about it & BARM the "IGP
computation side". We could do BAT & RAT for BIER & ROUTING but I fear that
the 2nd acronym is too unfortunate then and we don't capture the constraint
part either. So (BC,BA) and (RC,RA) will work fine as well IMO.

I see that in a sense as option F) with clear architectural framework and
am waiting for the writeup now to nit what's necessary ...

thanks

--- tony




On Wed, Feb 21, 2018 at 7:05 AM, Jeffrey (Zhaohui) Zhang <[e-mail]
> wrote:

> Hi Alia,
>
>
>
> Thanks for articulating it very clearly, and bring out the issue of having
> a clear specification on the interaction.
>
>
>
> I need one clarification from you though – is it possible that the you
> actually meant BART when you said BARM, and vice versa, in the following?
>
>
>
> ------------------
>
> With that, the BART can be exactly the IGP Algorithms registry.
>
> All the IGP Algorithms are available.   For the 2 currently defined, the
> constraints are empty.
>
>
>
> For the BARM, all the necessary constraints can be applied first.
>
>
>
> To define a code-point for BARM, the following information is necessary:
>
>    i) Constraints
>
>    ii) Can constraints be additive  (default yes unless specified
> otherwise)
>
>    iii) What is the algorithm - or is it "empty"
>
> -------------
>
>
>
> I’ll try to work with Les on the text to specify the interaction.
>
>
>
> Jeffrey
>
>
>
> *From:* Alia Atlas [mailto:[e-mail]]
> *Sent:* Wednesday, February 21, 2018 9:45 AM
> *To:* Jeffrey (Zhaohui) Zhang <[e-mail]>
> *Cc:* IJsbrand Wijnands (iwijnand) <[e-mail]>;
> [e-mail] <[e-mail]>;
> [e-mail]; IJsbrand Wijnands <[e-mail]>; [e-mail]; Eric
> Rosen <[e-mail]>
>
> *Subject:* Re: [Bier] BAR field length in draft-ietf-bier-isis-extensions
> and draft-ietf-bier-ospf-extensions
>
>
>
> First, I greatly appreciate the rapid education I have gotten on why the
> different aspects of this are important.
>
>
>
> Let us explore some details on the plan for an 8-bit BART and an 8-bit
> BARM that are independent.  Jeffrey,
>
> I really appreciate your bringing this option to the list.  It simplifies
> the option B idea of subtypes away and there
>
> seems to be a good amount of interest in it.
>
>
>
> My concern is that we specify adequately how the codepoints in BART and
> BARM interact.
>
>
>
> One of the challenges with doing so has been, IMHO, a bit of terminology
> where we are describing these as
>
> algorithms but in fact these are tuples consisting of a set of constraints
> and a base algorithm.
>
>
>
> BART is the BIER layer's constraints (BC) and algorithm (BA).  Let us
> describe this as BART =  (BC, BA).
>
>
>
> BARM is the Routing layer's constraints (RC) and algorithm (RA).  Let us
> describe this as BARM = (RC, RA).
>
>
>
> Let me give some concrete examples.
>
>
>
> Consider a case of BART=4 which is known to mean (prune non-BIER routers,
> use SPF).
>
> BARM = 200, which is known to mean (prune links with BW <10G, use SPF)
>
>
>
> It is desirable to have an outcome that is (prune non-BIER routers and
> links with BW < 10G, use SPF).
>
>
>
> This can work algorithmically because the constraints are the types that
> we've seen before for RSVP-TE.
>
>
>
> What I think we need to make the independent 8-bit BART plus 8-bit BARM
> work is a clear specification
>
> of the interaction. I think that is:
>
>
>
> Start with the topology Topology.
>
> 1) Apply the constraints represented by the BART  BC(Topology)
>
> 2) Apply the constraints represented by the BARM  RC(BC(Topology))
>
> 3) Select the algorithm A as follows:  use BA or if BA is "empty", use RA.
>
>     Run A on RC(BC(Topology)) to get the next-hops and information for the
> BIFT.
>
>
>
> Key points on the algorithm aspect are:
>
> a) BIER is the higher layer, so it is assumed to know better which
> algorithm should be used.
>
> b) It is possible for BA to be "empty"  (as the BARM=0 case discussed) so
> that the algorithm
>
>     falls through to whatever is RA.
>
>
>
> With that, the BART can be exactly the IGP Algorithms registry.
>
> All the IGP Algorithms are available.   For the 2 currently defined, the
> constraints are empty.
>
>
>
> For the BARM, all the necessary constraints can be applied first.
>
>
>
> To define a code-point for BARM, the following information is necessary:
>
>    i) Constraints
>
>    ii) Can constraints be additive  (default yes unless specified
> otherwise)
>
>    iii) What is the algorithm - or is it "empty"
>
>
>
> I am bringing forth this description now because I am concerned that the
> interaction between
>
> BARM and BART is not adequately defined to be published - but I see the
> strong interest in
>
> this as the solution and, from my understanding of the problem, agree.
>
>
>
> Please comment ASAP.
>
>
>
> Jeffrey & Les, is this something that you can turn into clear text for the
> drafts?
>
>
>
> Regards,
>
> Alia
>
>
>
>
>
>
>
>
>
> On Wed, Feb 21, 2018 at 9:17 AM, Jeffrey (Zhaohui) Zhang <
> [e-mail]> wrote:
>
> To make it absolutely clear using an example: even with <BART 1, BARM 200>
> it is still that the two fields are independent of each other.
>
> This particular combination means “apply BART 1” to “Flexible-Algo 200”,
> where “Flexible-Algo 200” could be “exclude red links”, while “BART 1”
> could be “skip BIER incapable routers”.
>
>
>
> This is a very practical and concrete example showing the advantage of
> having two separate fields. Other ways could be used to achieve the same
> result, but they’re more cumbersome.
>
>
>
> Jeffrey
>
>
>
> *From:* BIER [mailto:[e-mail]] *On Behalf Of *IJsbrand
> Wijnands (iwijnand)
> *Sent:* Wednesday, February 21, 2018 8:40 AM
> *To:* Jeffrey (Zhaohui) Zhang <[e-mail]>
> *Cc:* [e-mail]; [e-mail]; IJsbrand Wijnands <[e-mail]>;
> [e-mail] <[e-mail]>;
> Eric Rosen <[e-mail]>
> *Subject:* Re: [Bier] BAR field length in draft-ietf-bier-isis-extensions
> and draft-ietf-bier-ospf-extensions
>
>
>
>
>
>
>
> Ice: No, BART is not being slaved here. If BARM is 0, BART is all yours.
>
>
>
> Zzh> BART is BIER’s no matter what BARM is; not only when BARM is 0.
>
>
>
> Ice: Yes, sorry, I agree, BART is always BIER and BARM is always IGP.
>
>
>
> Ice: What I meant to clarify is that BART is not slaved to BARM (IGP) and
> v.s., if BART is used, BARM will just be 0.
>
>
>
> Thx,
>
>
>
> Ice.
>
>
>
>
>
> THx,
>
>
>
> Ice.
>
>
>
>
>
> Jeffrey
>
>
>
> Registry Algorithm a.k.a as BARM then ... Without this section we would be
> mandating that BARM is always an IGP algorithm or FA so basically it would
> mandate IGP
>
>
>
> Ice: Yes, BARM will be the IGP algorithm. That is to accommodate the
> people on the list who are of the opinion that aligning with IGP is
> important.
>
>
>
> Algorithm registry as the only option to perform a calculation making BART
> possibly pretty much useless ... Having a registry being mapped 1:1 into
> another registry known
>
>
>
> Ice: I don't understand why you are saying this. If BARM is 0, BART is all
> yours. Its unfortunate that a large part of the discussion is dominated by
> perceived functionality in the form of BIER Algorithm, while there is no
> architecture draft that describes how it should work and no discussion has
> happen in any IETF meeting, which leaves us all guessing. I think Alia
> asked a very good question on the list regarding "constraints". It is not
> at all clear if BART is a Algorithm or a Constraint. I think from your
> response you're saying its both, which seems wrong IMO.. To me Alia's
> question is still open, but that that may be because I could not decipher
> the rest of your response.
>
>
>
> as identity makes them both them the same thing by another name.
>
> So, to get anywhere close to consensus let's get bit less creative maybe
> and stick to the four letters of the alphabet that the AD extended as a
> wide playing field and the WG seems to converge around ... Or otherwise
> stick to option F) unmodified and see who's
>
> interested in it unless you insist on creating an option G) ...
>
>
>
> Ice: Jeffrey brought option F to the list in order to discuss it, that is
> what we are doing, and that is how you can converge on a solution and reach
> consensus. That is better compared to a vote on an option and everybody
> walks away with a different interpretation of it.
>
>
>
> Thx,
>
>
>
> Ice.
>
>
> _______________________________________________
> BIER mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/bier
> <https://urldefense.proofpoint.com/v2/url?u=https-3A__www.ietf.org_mailman_listinfo_bier&d=DwMFaQ&c=HAkYuh63rsuhr6Scbfh0UjBXeMK-ndb3voDTXcWzoCI&r=f7wsLGcfzAWDNS6XNTBZwj_OLAOsZZqdrR2IDAzeZqE&m=-s4yYMmN1jlRwgyk02_2IFamu-k7K7di1WiwKt-AoeE&s=pJHykwbjl-mkz5oRRkXE6hjOs0tsiC1ucjQtr6-F-4k&e=>
>
>
>
> _______________________________________________
> Isis-wg mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/isis-wg
>
>