Re: [Idr] BGP routes with color, Question 1: How does route resolution work with your feature?

[Natrajan Venkataraman <[e-mail]>]

Hi Dhananjaya,

This is Nats (One of the co-authors of the BGP-CT draft). I have questions on the following (NV)

    > >     Can there be more than one LCM present on a BGP CAR route?  If so, how does
    > >     that interact with resolution?

    I don't believe this question received an answer?

DR>> My bad. There can only be a single LCM present on a BGP CAR route.

    > >     > BGP CAR route resolution follows the same recursive resolution model that has been used in IP routing and MPLS networks for the past 20+ years (E via N). The only difference is that the route and next-hop are color-aware - (E, C) and (N, C).
    >
    > >     It's important to note that a property of this historical recursive model
    > >     for existing technologies is that it is longest prefix match.
    >
    > DR## Yes, and CAR does not attempt to change it.
    > DR## However, it is also worth noting that for technologies such as MPLS transport used historically for VPNs and other services, the recursive resolution ultimately has been a match on the host route for the PE/ASBR that’s the BGP service route’s next-hop. There is no aggregation. So, this point to some extent appears academic.

    The motivation for the above points is to illustrate that service route
    resolution is not solely (E,C) as received as the NLRI key.  The resolution
    key is (E, effective color).

DR>> Okay.

NV>> From the above,

I see that having the LCM attribute alone in the CAR NLRI universally solves the problem of communicating the color across different border nodes and towards the ingress PE nodes. All receivers of CAR NLRI need to make this local decision as to use the NLRI Key fully or to throw away the color in the key and use the LCM attribute as it creates an ambiguity where the NLRI key may be the same or different from the Resolution key. Are there any strong reasons for having color in the NLRI against having LCM as a sole representation of color?

Regards,
-Nats-


From: Idr <[e-mail]> on behalf of Dhananjaya Rao (dhrao) <dhrao=[e-mail]>
Date: Monday, March 21, 2022 at 6:32 AM
To: Jeffrey Haas <[e-mail]>
Cc: [e-mail] <[e-mail]>
Subject: Re: [Idr] BGP routes with color, Question 1: How does route resolution work with your feature?
[External Email. Be cautious of content]


Jeff,

Please see inline (DR>>)

On 3/21/22, 6:37 PM, "Jeffrey Haas" <[e-mail]> wrote:

    Dhananjaya,

    On Mon, Mar 21, 2022 at 10:35:22AM +0000, Dhananjaya Rao (dhrao) wrote:
    > On 3/20/22, 2:36 AM, "Jeffrey Haas" <[e-mail]> wrote:
    > >     "via (E,C)".
    >
    > >     When a Local-Color-Mapping Extended Community is present on the BGP CAR
    > >     route, does this mean that this CAR route contributes to resolution as
    > >     (E,LCM) rather than (E,C) where C is the original color present in the NLRI
    > >     key?
    >
    > DR## As you phrased it, LCM color is used when the route is used for resolution. Just to restate, there is no rewrite of the BGP NLRI color as it traverses hops.

    Thanks.  The property I'm trying to ensure we're clear on is that the RIB
    key, (E,C) is not necessarily the resolution key (E, effective color).  The
    effective color will be the C from the NLRI unless an LCM is present.

DR>> Yes. Section 2.8 of the BGP CAR draft states this along with the rationale.

    > >     If so, does that mean that when LCM is in use, the original (E,C) is not
    > >     permitted to contribute to resolution?
    >
    > DR## I did not understand the meaning of this comment. Can you elaborate ?

    The detail I'm trying to confirm is that if a BGP Speaker implementing -CAR
    has received a BGP route (E,C) with LCM C1 that it contributes to resolution
    solely as (E,C1) and not (E,C).

DR>> Okay. Hope the above reference clarifies the behavior.

    > >     Can there be more than one LCM present on a BGP CAR route?  If so, how does
    > >     that interact with resolution?

    I don't believe this question received an answer?

DR>> My bad. There can only be a single LCM present on a BGP CAR route.

    > >     > BGP CAR route resolution follows the same recursive resolution model that has been used in IP routing and MPLS networks for the past 20+ years (E via N). The only difference is that the route and next-hop are color-aware - (E, C) and (N, C).
    >
    > >     It's important to note that a property of this historical recursive model
    > >     for existing technologies is that it is longest prefix match.
    >
    > DR## Yes, and CAR does not attempt to change it.
    > DR## However, it is also worth noting that for technologies such as MPLS transport used historically for VPNs and other services, the recursive resolution ultimately has been a match on the host route for the PE/ASBR that’s the BGP service route’s next-hop. There is no aggregation. So, this point to some extent appears academic.

    The motivation for the above points is to illustrate that service route
    resolution is not solely (E,C) as received as the NLRI key.  The resolution
    key is (E, effective color).

DR>> Okay.

    > >     For cases where the only CAR prefixes in the system are of host length,
    > >     fall-back schemes for color seem fairly clear.  You would search on (N,C1)
    > >     and (N,C2) is permitted to match.
    > >
    > >     When the CAR routes are not host routes and there are perhaps overlapping
    > >     color fallback intents, e.g. if not found in C1 try C2, how should the
    > >     lookup functionally occur?  Would this be a longest prefix match vs. N for
    > >     C1 first and then fallback to a longest prefix match vs. N for C2?
    >
    > DR## That would be the functional behavior by default.
    > DR## I’m a bit unclear on whether you’re looking for anything specific in terms of behaviors or differences.

    The motivations for the two questions from the interim come from attempting
    to understand comparability between the two proposals.

    Thanks for helping to clarify this item.

    Once we have the missing answer for multiple LCM, I think we've closed the
    questions from the interim.  FWIW, Kaliraj indicates in -CT that it is a
    misconfiguration.  If the same holds for -CAR, that clarifies the intended
    behavior and perhaps suggests need for a section on the topic in each
    proposal's Error Handling section.

DR>> Ack, thanks for the perseverance.

Regards,
-Dhananjaya

    -- Jeff

_______________________________________________
Idr mailing list
[e-mail]
https://urldefense.com/v3/__https://www.ietf.org/mailman/listinfo/idr__;!!NEt6yMaO-gk!WY_ovfrLsfg2jcynx1JXcb7GKZ9PPEYNnY1F3ZJqc9V5P9d9wfyXSMIg4fMPVA$<https://urldefense.com/v3/__https:/www.ietf.org/mailman/listinfo/idr__;!!NEt6yMaO-gk!WY_ovfrLsfg2jcynx1JXcb7GKZ9PPEYNnY1F3ZJqc9V5P9d9wfyXSMIg4fMPVA$>


Juniper Business Use Only