Re: [Anima] Whether IPv6 only for ACP? //RE: I-D Action: draft-behringer-anima-autonomic-addressing-01.txt

[Duzongpeng <[e-mail]>]

Hi, Michael Behringer

Thanks for your reply. Please see inline.

Best regards
Zongpeng Du

-----Original Message-----
From: Michael Behringer (mbehring) [mailto:[e-mail]] 
Sent: Thursday, June 25, 2015 3:25 PM
To: Duzongpeng; Michael Richardson
Cc: Anima WG
Subject: RE: Whether IPv6 only for ACP? //RE: [Anima] I-D Action: draft-behringer-anima-autonomic-addressing-01.txt

OK, I think there is some confusion around this topic. 

A network with autonomic functions (we all agree that a fully autonomic network is not our main target for now) can support on the *data plane* any protocol the operator wants to use, including IPv4, appletalk and IPv8 (when it comes). 

[duzongpeng] Agree. [duzongpeng]

The statement is for the signalling, messaging and other protocols INSIDE the ACP, used by Autonomic Service Agents. 

Autonomic Service Agents will be exposed to a developer community, and the vendors implementing the ACP. There is absolutely no reason to standardise more than one protocol for this purpose. 

[duzongpeng] 
  I agree that it is vendor's job to implement ACP, and the ACP is kind of transparent for the operators. 

  Also, I agree that IPv6-based ACP works well for this purpose. 
  However, it does not mean that vendors can not implement an IPv4-based ACP.  

  Indeed, autonomic network's goal is to make management jobs as less as possible, but in some circumstance, the network operator do need to confirm every action of the network nodes. For examples, when debugging, or a high security level is suddenly needed. In this time, the ACP should not be transparent for the operators.

  We can not preclude this IPv4 requirement at the beginning. It should be the operator's option to use an IPv6-based or an IPv4-based one.
[duzongpeng]

MichaelR is right: You don't need to actively deploy IPv6 to use the ACP. It just happens. With one caveat: If you connect an NMS system into the ACP for "virtual out of band" usage, then that system needs to support IPv6. But that is typically not a problem, in our experience. 


[duzongpeng] 
  I agree that "it just happens". 

  However, if we have deployed an IPv4-based ACP, perhaps the solution for connecting an NMS system into the ACP for "virtual out of band" usage will become easier. Within an IPv4 network, perhaps the IPv4-based ACP has better compatibility than the IPv6 one.

  In my understanding, using IPv6 only ACP as you suggested will make the design jobs of autonomic control plane much easier. But, it is a tradeoff between doing less jobs and considering more about satisfying the different requirements of network managers. 

  Perhaps in future, when every network is IPv6, an IPv4-based ACP can be easily updated to IPv6-based because it can bootstrap itself, and all are autonomic.
[duzongpeng]

In other words: We're saying that *autonomic functions* (which the end user isn't exposed to at all) are exclusively IPv6, but the network can provide any service the operator wants. Specifically, an autonomic function (running on an IPv6 control plane) may well negotiate addressing for IPv4 services. 

Michael

> -----Original Message-----
> From: Duzongpeng [mailto:[e-mail]]
> Sent: 25 June 2015 04:06
> To: Michael Richardson; Michael Behringer (mbehring)
> Cc: Anima WG
> Subject: Whether IPv6 only for ACP? //RE: [Anima] I-D Action: draft- 
> behringer-anima-autonomic-addressing-01.txt
> 
> Hi, Michael Richardson:
> 
> 	I agree with your option. Perhaps ACP is not needed to be IPv6 only.
> 
> 	Of course, an IPv6 only autonomic control plane is ok for all the 
> services in autonomic network, and the solution has the advantage of 
> simplicity as mentioned by Michael Behringer.
> 
> 	However, the network operator may be more familiar with the IPv4- 
> base management.
> 	In the beginning of the ACP deployment, the network operator may not 
> be very confident about the decisions made by the autonomic network.
> Some kinds of monitoring jobs may be desired by the network operators. 
> At this time, the operator may want to see some IPv4 addresses. So, 
> perhaps we need to provide IPv4-based ACP as an option for these parts 
> of operators. Otherwise, we can only answer them "sorry, we only 
> support
> IPv6 here".
> 
> 	As IPv4 has been widely deployed, perhaps an IPv4-based autonomic 
> network will be accepted by more network managers at the beginning.
> 
> 	Indeed, at this time, an IPv4-based ACP is not as mature as the IPv6 
> one. Thus, perhaps some work needs to be done at this direction. For 
> example, as you mentioned, how can we get an address instead of the 
> ULA in IPv6.
> 
> Best regards
> Zongpeng Du
> 
> -----Original Message-----
> From: Anima [mailto:[e-mail]] On Behalf Of Michael 
> Richardson
> Sent: Thursday, June 25, 2015 1:12 AM
> To: Michael Behringer (mbehring)
> Cc: Anima WG
> Subject: Re: [Anima] I-D Action: draft-behringer-anima-autonomic- 
> addressing-01.txt
> 
> 
> Michael Behringer (mbehring) <[e-mail]> wrote:
>     >> > IPv6 only: Autonomic processes should use exclusively IPv6, for
>     >> > simplicity reasons.
>     >>
>     >> I agree with this as long as it's an RFC 2119 SHOULD, i.e. IPv4 
> could be used in
>     >> extraordinary circumstances. The GDNP design is fundamentally 
> IP- version
>     >> independent.
> 
>     > Yes, as long as it's really "extraordinary" ;-)   That 2119 says
>     > "SHOULD" means like you need a really good reason to not do it. What
>     > would such a good reason be? Just saying, in our implementation we
>     > often get the requirement "we want IPv4", and when you dig, the 
> reason
>     > is that they don't want to deploy IPv6, and that's the only
>     > reason. Slippery slope...
> 
> a) to respond to the immediate "we want IPv4" --- I'd just tell the customer
>    that it uses a custom link layer protocol that uses an ethernet type
>    different than 0x0800... just like CDP does :-)
>    (I don't think you need to "deploy" IPv6 to have an ACP...)
> 
> b) the only thing I care about is that nobody is allowed to say, "but
>    mechanism XYZ does not exist in IPv4".  That's all.  I don't know if
>    we will need IPv6-only things, but maybe.
> 
> ...
> 
>     >> Use-ULA: For these overlay addresses of autonomic nodes, we use 
> Unique
>     >> Local Addresses (ULA), as specified in [RFC4193].  An alternative
>     >> scheme was discussed, using assigned ULA addressing.  The consensus
>     >> was to use standard ULA, because it was deemed to be sufficient.
> 
> It might be that ULA is an IPv6-only feature.
> I don't consider 10/8 to be the same, as it is not statistically 
> unique inside the AS.
> 
> --
> Michael Richardson <[e-mail]>, Sandelman Software Works 
> -= IPv6 IoT consulting =-
> 
>