Le 7 avr. 2011 à 00:20, Lee, Yiu a écrit :

> Hi Remi,
> 
> I think I mis-read Figure 1 in your draft. As I read this, I think the
> current design will completely rely on dhcp to manage the v4 prefix. Yes,
> this decouples the v4 and v6 addresses, but this will add additional
> operation effort on dhcp. Also, if an ISP has many small IPv4 prefixes
> they want to use in 4rd, it will require to provision all of them to the
> BR. This may be ok for some operators, but not all operators.
> 
> In term of single or multiple 4over6 concentrators, I don't see people
> will advertise the same PI from two far away concentrators. So the return
> packet will also go back to the same (or close) concentrator. That said, I
> agree that stateless is always easier for HA. I have no disagreement on
> this.
> 

> In the end, they have pros and cons. Based on the ISP's requirements, it
> is just a matter of choice.

We agree.
Thanks.
RD

> 
> Cheers,
> /Yiu
> 
> On 4/6/11 5:19 PM, "Rémi Després" <[email protected]> wrote:
> 
>> Yiu,
>> 
>> I am not sure we completely understand each other here.
>> Please tell me if what I write is insufficiently clear.
>> 
>> Le 6 avr. 2011 à 21:06, Lee, Yiu a écrit :
>> 
>>> Hi Remi,
>>> 
>>> On 4/6/11 10:04 AM, "Rémi Després" <[email protected]> wrote:
>>> 
>>>> Partially agreed.
>>>> - If all IPv6 customers are given the same number of IPv4 ports, 4rd
>>>> doesn't imply anything on Pv6 routing.,
>>>> - If there are several 4over6 concentrators, there are constraints on
>>>> the
>>>> IPv6 routing to ensure satisfactory load sharing between them with
>>>> stable
>>>> routes.
>>>> - Furthermore, if these concentrators are far from one another, they
>>>> imply that the IPv4 backbone ensures symmetric routing so that all
>>>> return
>>>> traffic from a concentrator comes back to it.
>>> 
>>> I think we are on the same page. What I meant is in the case of stateful
>>> 6to4, IPv6 address won't contain any IPv4 address in the prefix. That
>>> won't require coupling between the IPv4 and IPv6 prefixes. It may or may
>>> not be a problem for some operators.
>> 
>> IPv6 prefixes of 6rd don't contain IPv4 embedded addresses (they differ
>> in this respect from the stateless A+P described in the aplusp draft).
>> With 4rd, if all CE's get IPv6 prefixes having the same length, and are
>> assigned the same number of IPv4 ports, the IPv6 routing is taken as is.
>> It is only after that that one or a few mapping rules, whose parameters
>> can be advertised in stateless DHCPv6, define how each CE derives its
>> shared IPv4 address and port set from its IPv6 prefix.
>> 
>>> LB is a different issue. It doesn't relate to address coupling. That
>>> said,
>>> it isn't so straight forward for stateless approach because it depends
>>> on
>>> the placement of the BR and the v6 routing to the anycast address.
>> 
>> Routing of the IPv6 network can be reconfigured at any time without
>> breaking connections.
>> Load balancing can be performed without any concern for packets of each
>> connection having to go to the same BR, which makes it simpler.
>> 
>> 
>>> For stateful 4to6, the outgoing and incoming v4 traffic are symmetric
>>> because all v4 traffic must traverse the concentrator. I don¹t see why
>>> the
>>> backbone has to do something special for it.
>> 
>> If there is only one concentrator, no problem.
>> But if there are several widely separated concentrators, and they are
>> reachable at the same PI prefix at the border between the IPv6-only
>> network and the IPv4 backbone, then there is a return path problem.
>> 
>> RD
>> 
>> PS: I may be able to continue tomorrow, but then will be unreachable
>> until April 25.
>> 
>> 
>> 
>> 
> 


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

Reply via email to