Re: [bess] draft-ietf-bess-evpn-overlay-10 PMSI with Ingress Replication

[Thomas Morin <[e-mail]>]

Hi Don,

Fedyk, Don, 2017-12-14 20:33:
> I think the gray area is that this draft talks about BUM traffic and
> ingress replication and then has a section on Multicast tunnels which
> excludes ingress replication traffic from the tunnels. 

No, ingress replication is not excluded at all:

   The following tunnel types as defined in [RFC6514] can be used in
   the PMSI tunnel attribute for VXLAN/NVGRE:

         + 3 - PIM-SSM Tree
         + 4 - PIM-SM Tree
         + 5 - BIDIR-PIM Tree
         + 6 - Ingress Replication

> If you are using point to point VXLAN/NVGRE  tunnels then ingress
> replication is default [...]

This formulation surprises me: that some implementations behave as you
describe is possibly true (this seems to be the case of the
implementation that triggered this discussion), but I don't know about
any text in the specs we are discussing that would imply such a
'default'.  

You might have implementations that in the absence of any local
configuration for an EVPN instance on which type of tunnel to use for
BUM, will default to ingress replication: this is fine, out of the
scope of what is specified for interop, and not breaking other
implementations (as long, of course, that what is chosen locally is
then advertised as expected in a PMSI Tunnel Attribute).


> but IMET is being used to identify the NVE IP.    I read RFC7432 and
> RFC6514 in this area and thought that the PMSI attribute MUST be set
> when there is an Inclusive Multicast Ethernet tag IMET.  

Yes!  (the text of RFC7432 quoted by Ali reminds us that)

 
> I can see two possible fixes:
> -          Specify that the PMSI attribute MUST be set if there is an
> IMET route and specify correct attribute. 

Given the content of RFC7432 and the fact that this is a normative ref
of draft-ietf-bess-evpn-overlay, I think that we don't need to repeat
this MUST in draft-ietf-bess-evpn-overlay.  That is, unless we
explicitly identify an ambiguous piece of text.

> -          Allow that ingress replication is default when PMSI is
> absent but accept PMSI that specifies ingress replication.
> 

I don't think we should do that. It would overnight make non-compliant
pre-standard implementation of draft-ietf-bess-evpn-overlay, without a
rationale to do so except coping with an implementation that assumed a
bit too much.

Best,

-Thomas


 
> From: BESS [mailto:[e-mail]] On Behalf Of Marco Marzetti
> Sent: Thursday, December 14, 2017 9:21 AM
> To: Thomas Morin <[e-mail]>
> Cc: [e-mail]
> Subject: Re: [bess] draft-ietf-bess-evpn-overlay-10 PMSI with Ingress
> Replication
>  
> Hello,
>  
> I have encountered an implementation that is not attaching any PMSI
> to the IMET.
> The authors think they don't really need it because they only support
> Ingress Replication.
> Such behavior breaks interoperability with other implementations that
> are dropping the NLRI if PMSI is not attached.
>  
> So i looked at draft-ietf-bess-evpn-overlay-10 and noticed that
> there's no clear indication of what the proper behavior is.
> As said i assumed i had to look at RFC7432 and RFC6514 (and i did
> it), but i wasn't 100% sure and i preferred to ask.
>  
> Onestly you already made my day by confirming what i thought.
> My suggestion was to make things more clear, but i admit that it
> could look redundant.
>  
> Thanks
>  
>  
>  
>  
>  
>  
>  
> On Thu, Dec 14, 2017 at 2:39 PM, Thomas Morin <[e-mail]
> m> wrote:
> > Hi Marco,
> > 
> > Marco Marzetti, 2017-12-14 12:25:
> > > I am writing this email asking you to clarify what's the
> > suggested
> > > behavior when PMSI Tunnel Type is set to "Ingress Replication"
> > (type
> > > 6) as draft-ietf-bess-evpn-overlay-10 only suggests what to do
> > with
> > > multicast tunnel trees.
> > >
> > > I think the originating PE should conform with RFC6514 and
> > RFC7432
> > > (from which you've taken inspiration) and always (RFC2119 MUST)
> > > attach PMSI Tunnel attribute with the Tunnel Type set to Ingress
> > > Replication and Tunnel Identifier set to a routable address of
> > the PE
> > > itself (more specifically NVE's IP address).
> > >
> > > Is that correct?
> > > In that case i suggest to add the following line at the end of
> > > Section 9.
> > > """
> > > For Ingress Replication the PE should follow what's stated in
> > RFC6514
> > > Section 5 .
> > > """
> > 
> > The text of section 9 lists "Ingress Replication" in the list of
> > tunnel
> > types that can be used. My understanding is that, in the absence of
> > anything being specifically said for Ingress Replication, an
> > implementation should follow what is said in RFC7432 and RFC6514.
> > (What
> > other specs could it follow to implement this supported type ?
> > RFC7432
> > and RFC6514 are more than an inspiration here, these are specs that
> > the
> > document refers to explicitly)
> > 
> > So I'm not sure that it is useful or needed to add text.
> > 
> > Can you perhaps expand on why the current text would possibly be
> > ambiguous, misleading or incomplete...?
> > 
> > -Thomas
> > 
> 
> 
>  
> --
> Marco