Re: [RTG-DIR] RtgDir review: draft-kompella-l2vpn-l2vpn-07.txt

[Bhupesh Kothari <[e-mail]>]

Bocci, Matthew (Matthew) <[e-mail]> wrote:

> Comments:
> 
> Major Issues:
> 
> - Relationship to published and on-going work in the L2VPN and PWE3 working
> groups. This draft describes an alternative to the procedures for signalling PW
> labels used in pt-pt layer 2 VPNs (VPWS) and for auto-discovery of endpoints
> participating in a L2VPN. The existing IETF-approved technologies use LDP for
> the signalling (as described in RFC4447) and a combination of that with BGP for
> auto-discovery (RFC6074). Since this draft describes an alternative that was
> rejected by the L2VPN WG, I would suggest adding a disclaimer to the abstract
> and introduction, as per the recommendation of RFC5742, along the lines of "The
> IETF technology for this function is specified in RFC6074 and
> RFC4447",

Is this a blocker for this draft to get published?  If not, I would
like to not include your suggestion.

Or, to be fair, can I add the following:

"
The IETF techonology for this function is specified in RFC6074 and
RFC4447 but following are the disadvantages of each:

<disadvantages>
"

I don't want a reader to assume that the IETF "approved" technology
equals "better" technology.  There are pros and cons of each, and SPs
are the ultimate judge of which solution meets the requirements of
their networks.  In this case, this is 8/10 year old technology with
many SP adoptions, even after IETF decided to not purse it.

In addition, this is an Informational RFC and so it is clear to
everyone that IETF did not adopt it for standards track.  What more is
required?



> in
> addition to the suggestions on the list that implementations of this draft must
> also comply to these RFCs.  

What is the rationale for this?

Stating this adds no value, increases unnecessary complexity for both
vendors and customers and is irrelevant, at least from a technical
point of view.

Neither RFC that you are mentioning is required for proper functioning
of this draft.  Why would you like to force a vendor to implement
complete disjoint techonologies?  As a result, you are also forcing
the provider to incur the cost of extra code which is not at all
needed if he/she wishes to deploy this techology.

Also note that in all existing deployment cases, the routers that have
implemented this draft also supports RFC4447.

> 
> - Section 9: IANA Considerations. This section requests a registry be created
> with standards action code points. I believe this is not appropriate as this is
> an informational draft. There has already been some discussion on this point on
> the list, and that is noted as a part of this review.


This has been discussed.  Conclusion is to replace 'standards action'
with 'expert review'.



> 
> - Section 5: Layer 2 VPN interworking. This section introduces a layer 2
> transport service for IP packet. However, just encapsulating IP packets between
> PEs will not be sufficient for a layer 2 VPN service between the CEs. You also
> need to support something to allow L2 address resolution between the CEs, for
> example translation of the neighbour discover protocols (e.g. ARP over Ethernet
> and InvARP over FR). Since this is describing the L2VPN, and not just the PW
> encapsulation, I think this section should either say how this is done with BGP
> (draft-ietf-l2vpn-arp-mediation uses LDP) or just say it is out of scope of
> this particular draft and the L2 address is statically configured.


I am OK with 'out of scope ...' statement. 


> 
> 
> 
> Minor Issues:
> 
> General:
> - The draft mainly discussed pt-pt L2VPNs. The general term for this is VPWS,
> but this term is not mentioned anywhere. It would help to clarify with a
> statement in the introduction that this the subject of the draft.


Agree.



> 
> - It would help if the figures showing the TLV structures showed the detailed
> structure, throughout, including bit fields, etc, in line with common practice
> in IETF RFCs.


There is only one TLV defined in the draft (Table 2).  Can you point
which other places need changes?


> 
>  
> 
> - 1. Introduction, 5th paragraph:
> "Technically, the approach proposed here uses the concepts and
>    solution and described in [RFC4761], which describes VPLS, a
>    particular form of a Layer 2 VPN."
> 
> This reads as if RFC4761 is *the* description of VPLS. This is not the case and
> the IETF has standardised two VPLS methods. I propose modifying this to "…which
> describes a method for VPLS,…".


Agree.


> 
> - 1.1 Terminology. Final paragraph.
>    "These will be referred to as Attachment
>    Circuits (ACs), following [RFC4447].  Similarly, the entity that
>    connects two attachment circuits across the Service Provider network
>    is called a pseudo-wire (PW)."
> 
> RFC3985 is probably the correct reference.
> Also: 
>  s/pseudo-wire/pseudowire


Agree.



> 
> - 1.2.5: PE Scaling
> This section spends a lot of time discussing the scaling issues of L3 VPNs. I
> am not sure why this is required in a BGP L2VPN draft, unless the issue is that
> because the control plane is the same, many of the same scaling considerations
> apply. Please can you clarify if this is true, and if so make such a note in
> the text?

The intention of this sub-section is to highlight the better scaling
properties of BGP VPWS in comparision to BGP L3 VPNs.  In order to
highlight, the sub-section has a couple of paragraphs on what the
L3 VPN scaling issues are on PEs and RRs that also hold the BGP VPWS
state. 



> 
> - 1.3: Advantages of Layer 3 VPNs
> I do not think this adds any value to a draft on L2 VPNs. The content may be
> useful elsewhere, but I think it would be better if this was removed from this
> draft. 


The content is useful.  It is also a bit historic in the sense that
when this draft came out (8/10 years back), it was more appropriate to
contrast L2 and L3 VPNs.

I am OK if this section was removed. 


> 
> - 3: Operation of Layer 2 VPNs
> "In what follows, Frame Relay serves as the Layer 2 medium, and each
>    CE has multiple DLCIs to its PE, each to connect to another CE in the
>    VPN.  If the Layer 2 medium were ATM, then each CE would have
>    multiple VPI/VCIs to connect to other CEs. "
> 
> I think I know what you mean by "medium" I.e. The underlying link layer, where
> each group of DLCIs is an AC in the RFC3985 sense, plus DLCI acts as a
> demarkation of a L2VPN service between the PE and CE. If so, I think it would
> help to add this clarification.


Agree. 




> - 6.3: IP-Only Layer 2 Interworking:
> 
>          +----------------------------+
>          | Tunnel |  VPN  |    IP     |     VPN label is the
>          | Encap  | Label |  Packet   |     demultiplexing field
>          +----------------------------+
> 
>           Figure 3: Format of IP-only Layer 2 interworking packet
> 
> I think the field marked "Tunnel Encap" should instead be labelled "PSN
> Transport Header" or "MPLS Transport Header", in common with the other
> encapsulation RFCs that you have referenced.
> 


Agree.



> 
> Best regards
> 
> Matthew
> 


Thanks for the review.


Bhupesh