Thomas, 

Since there are quite a few people stating that NVE-NVE control plane is needed 
for NVO3, the architecture document should at least have a sub-section 
describing the trade-off of having NVE-NVE protocol. 

Even with one TS being connected to one NVE, there are people saying that one 
NVE could have two different uplink ports to underlay network, therefore might 
have two different IP addresses. My counter-argument is that NVE should use its 
loopback IP address and can have different MAC addresses for its different 
uplink ports or use NIC-Teaming for its multiple ports. 

I think that my suggested subsection is necessary to describe the trade-off and 
to make WG feel that their comments are well addressed.  

Granted that NVE-NVE protocol does provide some value, like notifying each 
other the directed attached TSes when NVA is not reachable, etc. 
NVO3 architecture doesn't have to "require this protocol", but doesn't have to 
" explicitly exclude this protocol " either.  

Again here is my suggested new subsection (with minor change to address 
Pankaj's comments:
-----------------------------------

4.4 The trade-off of Inter-NVE control plane protocol (Suggested new 
sub-section) 

There could be various reasons, link failure, node failure, or others, causing 
egress NVEs (or even NVA) not reachable. Without any NVE<->NVE control plane 
protocol, the ingress NVE is not aware of the reachability of egress NVE 
causing encapsulated packets to dropped somewhere in the underlay network. 

If most TSes are only attached to a single NVE and traffic to NVEs are not 
aggregated flows, then the value provided by NVE-NVE control plane protocol is 
not significant compared to the extra cost added to NVE.  When one NVE has 
multiple uplink ports to the underlay network, it is recommended to use NVE's 
loopback address instead of port addresses to avoid the complication on ingress 
node in determining which address to use. 
Under this environment, the difference between having “NVE-NVE control 
protocol" is whether the packets being dropped at the Ingress NVE or somewhere 
in the underlay network when egress NVE is not reachable.  
In data center environment where most communications are among applications, 
most likely a source will not send more packets without acknowledgment from the 
destination. Then the impact of where the packet is dropped is not that big. 
However, in an environment where TSes are connected to multiple NVEs or high 
chance of connection failure between NVE & NVA, then it might be worthwhile to 
consider the Inter-NVE control plane protocol, so that ingress NVE can choose 
different egress NVE for a given target.

---------------------------

Linda

> -----Original Message-----
> From: Thomas Narten [mailto:[email protected]]
> Sent: Wednesday, November 20, 2013 11:07 AM
> To: Linda Dunbar
> Cc: [email protected]
> Subject: TS conects to VN through one NVE [was Re: [nvo3] No need for
> NVE-NVE control plane [was Re: Poll for WG adoption and IPR check for
> draft-narten-nvo3-arch-01.txt]
> 
> Linda,
> 
> > If most TSes are only attached to a single NVE and traffic to NVEs
> >  are not aggregated flows, then the NVE-NVE control plane protocol
> >  doesn���t provide much benefits.
> 
> I want to flag the first part of the sentence above because this
> question has been raised before. It suggests that a TS might connect
> to more than one NVE at a time.
> 
> From an architectural perpsective, I think we should say "no". Life is
> much simpler if a TS connects to a VN through exactly one NVE. More
> precisely, a TSI always connects to one NVE. If we allowed connections
> to multple NVEs simultaneously, it would raise all sorts of questions
> as to when to use one over the other, etc.
> 
> I don't think I'm actually saying anything new with this... I've
> always assumed this, documents are written with this in mind, and it
> hasn't come up (other than in passing) on the list.
> 
> A TS can still connect to multiple different VNs through different
> network interfaces, and hence different TSIs, but I think we should
> have a TSI always connect to exactly one NVE.
> 
> Any other opinions?
> 
> Thomas
> 

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

Reply via email to