Re: [bess] bess: draft-zzhang-bess-mvpn-msdp-sa-interoperation

["Jeffrey (Zhaohui) Zhang" <[e-mail]>]

Hi Toerless,

> -----Original Message-----
> From: BESS [mailto:[e-mail]] On Behalf Of Toerless Eckert
> Sent: Friday, November 24, 2017 1:31 AM
> To: Lenny Giuliano <[e-mail]>
> Cc: MBONED WG <[e-mail]>; Toerless Eckert <[e-mail]>;
> [e-mail]
> Subject: Re: [bess] bess: draft-zzhang-bess-mvpn-msdp-sa-interoperation
> 
> Hi Lenny,
> 
> a) I was asking Jeffrey in the WG meeting if he saw any difference
>    between MSDP and 4610 wrt. to the applicability of your draft, and
>    he also said no (as i think too), thats where this originated.

On the spot I did make a mistake that I thought with 4610 the RPs would propagate registers (received from a C-RP) to another :-) Lenny caught that.

> 
> b) Wrt to your contentions below:
> 
>    1) I agree, #1 is covered - for both MVPN SA and rfc4610 register
>    messages according to rfc6514, 14.1.
> 
>    2) Let first make the argument where MSDP and rfc4610 operate equally:
>    when you configure an MSDP mesh group. If all the non-PE RPs in an
>    MVPN C domain are either an MSDP mesh group or an rfc4610 RP-set
>    (same function, different names), then you could just add some PE
>    as members of the MSDP mesh group or the RFC4610 RP-set and that
>    would work "fine" without extending rfc6514 14.1, aka: those PEs
>    would learn active C (S,G) and distribute it to all other PE, and
>    they would not need to send any (S,G) SA/register messages back to the
>    non-PE mesh group/rp-set because those RPs already have all state
>    they need. And, as your draft says, as soon as you add more than
>    one PE this way, you have multiple PEs that will send to all other
>    PEs all active C (S,G) state info. And its kinda difficult to figure
>    out how much redundancy one wants to set up that way.

Correct.

> 
>    So, the model that i thought your draft meant to establish is one
>    where you would avoid having to build a mesh-group/RP-set across all
>    non-PE RPs in a large network:

Not to "avoid" -- just to make it work (better) -- when there is no such mesh-group/RP-set across all non-PE RPs, by making all RP-PEs into an MSDP mesh-group of their own (though replacing MSDP session with BGP-MVPN session).

> 
>    You would logically have a per-MVPN-site mesh-group/RP-set between
>    the PEs of that site (one or more) and one or more non-PE RP.

And the RP-PEs will be in their own "core" mesh.

>    Whenenver a PE receives an MVPN SA that comes from another site,
>    it would forward it to the other non-PE members of this mesh-group/RP-set,
>    and your new draft section 3 would permit this type of source check
>    (i am not completeley through understanding that part of your draft).
> 
>    So, that would be a very scaleable setup as soon as you have
>    non-PE RPs in the MVPN.
> 
>    3) MSDP of course has setup options not shared with rfc6410, aka:
>    any setup that is more than a mesh-group. If you intended for the
>    draft to also apply to such cases,

Yes.

>    it would be great to give
>    examples of the so tht one can check if/how the proposed spec in your
>    draft would work in the fce of those cases existing MSDP RPF rules
>    (which can of course be quite complex).

Yes more text would help. The idea is that MVPN SA messages would just be treated as MSDP SAs from the PE that originated the SAs. Those MSDP SAs received from true MSDP peers and those treated as MSDP SAs would follow the existing MSDP rules.

> 
> 3) Btw: It would be good if you would add some picture to your draft
>    starting from section 2 to illustrate your text example PE1, PE2
>    and so on.

Sure.

> 
>    Also, your draft says:
> 
>      [RFC6514] only specifies that a PE receiving the MVPN SA routes, say
>      PE2, will advertise (C-S,C-G) C-multicast routes if it has
>      corresponding (C-*,C-G) state learnt from its CE
> 
>    can you point me to the text in 6514 that says that ?

The full text is the following:

14.2.  Receiver(s) within a Site

   ...

   When (as a result of receiving PIM messages from one of its CEs) a PE
   creates, in one of its MVPN-TIBs, a (new) (C-*,C-G) entry with a non-
   empty outgoing interface list that contains one or more PE-CE
   interfaces, the PE MUST check if it has any matching Source Active
   A-D routes.  If there is one or more such matching routes, and the
   best path to C-S carried in the matching route(s) is reachable
   through some other PE, then for each such route the PE MUST originate
   a Source Tree Join C-multicast route.  If there is one or more such
   matching routes, and the best path to C-S carried in the matching
   route(s) is reachable through a CE connected to the PE, then for each
   such route the PE MUST originate a PIM Join (C-S,C-G) towards the CE.

   When, as a result of receiving a new Source Active A-D route, a PE
   updates its VRF with the route, the PE MUST check if the newly
   received route matches any (C-*,C-G) entries.  If there is a matching
   entry, and the best path to C-S carried in the (A-D) route is
   reachable through some other PE, the PE MUST originate a Source Tree
   Join C-multicast route for the (C-S,C-G) carried by the route.  If
   there is a matching entry, and the best path to C-S carried in the
   (A-D) route is reachable through a CE connected to the PE, the PE
   MUST originate a PIM Join (C-S,C-G) towards the CE.

Jeffrey

> 
> Cheers
>     Toerless
> 
> On Wed, Nov 22, 2017 at 02:38:09PM -0800, Leonard Giuliano wrote:
> > Toerless,
> >
> > Thanks for the comments.  After thinking about your feedback on RFC4610 a
> > bit, I'm not sure that case is applicable here.  Consider the 2 directions
> > of interworking:
> >
> > 1) MSDP SA/AnycastRP_PIM_Register -> MVPN SA
> >
> > 2) MVPN SA -> MSDP SA/AnycastRP_PIM_Register
> >
> > As I understand, #1 is already covered by RFC6513/6514.  #2 is the missing
> > piece that this draft attempts to address.  I don't believe #2 will be
> > applicable for RFC4610, as these registers only go to members of the RP
> > set.  And the RP set should be configured on all the C-RPs.  It wouldn't
> > make much sense to have these registers transit the MVPN domain to go to
> > an AnycastRP not in the configured RP set.
> >
> > Hope this is clear, and let me know if I'm missing anything.
> >
> > -Lenny
> >
> > | -----Original Message-----
> > | From: BESS [mailto:[e-mail]] On Behalf Of Toerless Eckert
> > | Sent: Monday, November 13, 2017 9:43 PM
> > | To: [e-mail]
> > | Cc: [e-mail]
> > | Subject: [bess] bess: draft-zzhang-bess-mvpn-msdp-sa-interoperation
> > |
> > | Jeffrey presented subject draft in mboned. Given how i am
> > | not usually tracking BESS WG mailing list and may not be around:
> > |
> > | I would like to see subject draft to be adopted as a WG document in BESS
> > | and become an update to RFC6514 (not to say bugfix ;-).
> > |
> > | Feeedback detail: The draft should be amended to fix the same problem
> not only for
> > | MSDP SA but also RFC4610 and probably accordingly change the draft
> name.
> > |
> > | Cross-posted to mboned (sorry) because there where a couple of MBoned
> > | participants expressing support for the draft in BESS and may like me not
> > | be BESS regulars.
> > |
> > | Cheers
> > |     Toerless
> 
> --
> ---
> [e-mail]
> 
> _______________________________________________
> BESS mailing list
> [e-mail]
> https://urldefense.proofpoint.com/v2/url?u=https-
> 3A__www.ietf.org_mailman_listinfo_bess&d=DwICAg&c=HAkYuh63rsuhr6Scbf
> h0UjBXeMK-
> ndb3voDTXcWzoCI&r=f7wsLGcfzAWDNS6XNTBZwj_OLAOsZZqdrR2IDAzeZqE&
> m=hQz-NA6r7ahWKqMy6aI7ZXX5lvGX1ubRovr3Co5_VNE&s=rC5k8JOSHYD-
> XaNaPGxhtl-yyDQhzqd909XjwXnS5EI&e=