Hi Zu,

My responses are inline below.

 - Larry

From: Zu Qiang <[email protected]<mailto:[email protected]>>
Date: Thursday, September 12, 2013 9:04 AM
To: Larry Kreeger <[email protected]<mailto:[email protected]>>
Cc: "[email protected]<mailto:[email protected]>" <[email protected]<mailto:[email protected]>>
Subject: Questions to draft-ietf-nvo3-nve-nva-cp-req-00

Hello, Larry
                I have some clarification questions on section 3 of this draft.

-          Section 3.3: the text is a little bit confused. Please clarify it.

1)      At VN connect/disconnect, the attached/detached NVE may be notified by 
the hypervisor. Then the NVE will inform the NVA, where the NVA is going to 
update the remote NVEs with a new address mapping table.

2)      Or the NVA may learn the VN connect/disconnect from somewhere (e.g. 
orchestration). Then it will notify the NVE where the VN was attached/detached 
and update the remote NVEs with a new address mapping table.

3)      Or both 1) and 2) are allowed?

LK> In my opinion, number 1) makes the most sense.  Doing thing this way 
eliminates the need for a central orchestrator that must manage all NVEs 
regardless of where they might reside (hypervisors, ToR switches, routers, 
firewalls, load balancers etc).  However, section 6.1 of draft-marten-nvo3-arch 
leaves it open for number 2) as well.  As long as we have a protocol to 
federate NVAs, one might have one NVA only for hypervisors with embedded NVEs 
where VN connect/disconnect is communicated directly to the NVA (perhaps 
through an internal software interface), and another NVA for various physical 
NVEs.


-          Because of my confusion from the reading of section 3.3, it may be 
good to make it more clear in section 3.1 and 3.2 on which direction the 
notification is sent. For instance, “A protocol is needed for the NVA to 
provide this inner to outer mapping and VN Context to each NVE that requires it 
and keep the mapping updated in  a timely manner.”

LK> Yes, this section is meant to express the need for an NVE to get a mapping 
from the NVA.  I can add the text you suggest above to make it clearer.  
However your comment made me realize that the other direction (NVE telling the 
NVA what inner addresses to outer mappings it terminates) is missing.  I can 
add subsections to 3.1 to cover each direction of inner/outer mapping 
information flow.


-          Section 3.4: I agree that VN name into VN ID mapping may be needed. 
However, we may need to allow different implementations. For instance, one can 
have this mapping in NVA only and use VN ID in all interfaces. Therefore, 
supporting this mapping in NVA-NVE protocol shall be optional.

LK> I agree that it is optional.  The wording in section 3.4 was trying to 
convey this.  It mentions both approaches. It is probably better to come out 
and say this is optional instead of inferring it with statements such as "Thus, 
it may be useful for the NVE-to-NVA protocol to support an operation that maps 
VN Names into VN IDs.".

Have a nice day
Zu Qiang

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

Reply via email to