Hi Larry,

Thanks fo your thorough review.
See my comments inline.

Marc 

> -----Original Message-----
> From: [email protected] [mailto:[email protected]] On 
> Behalf Of Larry Kreeger (kreeger)
> Sent: Tuesday, October 30, 2012 1:55 AM
> To: [email protected]
> Subject: [nvo3] Comments on draft-bl-nvo3-dataplane-requirements-02
> 
> Hello draft-bl-nvo3-dataplane-requirements authors,
> 
> I was reviewing this latest 02 draft fully with a special 
> focus on all the "MUST" statements and have the following comments:
> 
> 1) In Section 3.2.1 it says:
> "An L2 VNI MUST provide an emulated Ethernet multipoint 
> service as if Tenant Systems are interconnected by an 802.1Q 
> LAN over a set of NVO3 tunnels."
> 
> Does this mean that the inner Ethernet frame must carry an 802.1Q tag?
> Maybe it can be reworded to make it clearer by explicitly 
> saying something about VLAN tags.

[ml] How about "An L2 VNI MUST provide an emulated Ethernet multipoint 
service as if Tenant Systems are interconnected by an bridge over a set of NVO3 
tunnels.
The bridge may be 802.1Q VLAN enabled."?

> 
> 
> 2) In Section 3.2.1 it says:
> "Loop avoidance capability MUST be provided."
> 
> What does loop avoidance mean in the data plane? Is split 
> horizon forwarding enough? Is this implying that a TTL is 
> needed in the encapsulation header (inside the outer IP header)?

[ml] No assumption is made. Either mechanism is applicable.

> 
> 3) In Section 3.2.1 it says:
> "As frames arrive from VAPs or from overlay tunnels, the MAC 
> learning procedures described in IEEE 802.1Q are used:..."
> 
> Is there a more specific reference to a part of 802.1Q that 
> describes inner source MAC address to outer source address 
> learning? It may confuse readers because this draft refers to 
> inner MAC to outer L3 address.  Is the reference to 802.1Q required?

[ml] How about, "..., standard MAC learning procedures are used:"?

> 
> 
> 4) In Section 3.2.1 it says:
> "Broadcast, Unknown Unicast and Multicast (BUM) traffic 
> handling MUST be supported."
> 
> It may be desirable for an operator to have the ability to 
> disable unknown unicast (if there is a control plane to 
> distribute MAC to L3 tunnel mappings), and also to disable 
> multicast if it is a service that they do not want to support.
> Are you saying that an implementation must always forward 
> unknown unicast and multicast, or that it must be "capable" 
> of forwarding them if desired?

It is the latter. The text will be clarified in the next revision.

> 
> 
> 5) In Section 3.2.1 it says:
> "To achieve this, the NVE MUST support ingress replication 
> and MAY support multicast over an overlay multicast tree."
> 
> I don't see why ingress replication is a MUST if the underlay 
> can support multicast.  I think this should be reworded to be 
> "To achieve this, the NVE MUST either support ingress 
> replication or MUST support multicast across the underlay 
> network if the underlay supports it."

[ml] Agreed. The text will be changed accordingly.

> 
> 6) In Section 3.3.2 it says:
> "From an encapsulation perspective the IPv4 and IPv6 
> encapsulations MUST be supported, MPLS tunneling MAY be supported."
> 
> Requiring support for BOTH IPv4 and IPv6 seems too strong.  
> Can this be changed to "MUST support IPv4 or IPv6 and should 
> support both, MPLS tunneling MAY be supported."

[ml] Agreed.

> 
> 7) In Section 3.3.2.1 :
> Is this mandating a specific entropy field in the 
> encapsulation header, or would using the source UDP port or 
> the GRE key qualify?  If they would qualify, it would be good 
> to add a sentence to say so.

Both are applicable. The text will be changed to reflect this.

> 
> 8)In Section 3.3.2.3 it says:
> "L2 NVEs MUST support ingress replication and SHOULD support 
> point-to-multipoint tunnels."
> Must seems strong for ingress replication.  There should be a 
> MUST support ingress replication or point-to-multipoint 
> tunnels. Also, this is inconsistent with section 3.2.1 where 
> p2mp tunnels are not mentioned but instead mentions an 
> overlay multicast tree. Does a multicast tree qualify as a 
> p2mp tunnel ?

[ml] Correct, it MUST support either ingress replication or multicast (e.g. IP 
multicast trees or p2mp tunnels).
The text will be changed accordingly.

> 
> 9) General comment: In a few places there are instances of 
> "must" and "may" that are not capitalized.

This will be fixed.

> 
> Thanks, Larry
> 
> _______________________________________________
> nvo3 mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/nvo3
> 
_______________________________________________
nvo3 mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/nvo3

Reply via email to