Le 2012-02-07 à 16:24, Satoru Matsushima a écrit : > On 2012/02/07, at 16:10, Rémi Després wrote: > >> >> Le 2012-02-07 à 15:26, Satoru Matsushima a écrit : >> >>> On 2012/02/07, at 14:03, Rémi Després wrote: >>> >>>> >>>> Le 2012-02-07 à 13:07, Satoru Matsushima a écrit : >>>> >>>>> Hi Remi-san, >>>>> >>>>> On 2012/02/07, at 11:13, Rémi Després wrote: >>>>> >>>>>> Hello Ole, Tetsuya-san, Wojciech, >>>>>> >>>>>> In a use case described in the 4rd-U draft (sec 5.3), an ISP replaces >>>>>> its dual-stack routing by IPv6-only routing. >>>>>> For this, independently from the number of IPv4 prefixes it has to >>>>>> support, it uses only one mapping rule. >>>>>> (By replacing each IPv4 route by an equivalent IPv6 route, it ensures >>>>>> that all customers keep their IPv4 addresses.) >>>>>> >>>>> >>>>> I don't think that it could work as you explained in that section. For >>>>> example, the BR would need to check a received packet from a CE whether >>>>> it has correct source address in mapping rule or not. It means that the >>>>> BR must know all address mappings for CE between IPv4 addresses and IPv6 >>>>> prefixes. Is it correct understanding? >>>> >>>> Ingress filtering of the domain has checked that the IPv6 source starts >>>> with the delegated IPv6 prefix, a /112 which includes the IPv4 address. In >>>> the 4rd-E case, the BR checks that the source address in the IPv4 header >>>> matches that of the IPv6 address. There is therefore no need for the BR to >>>> know all IPv4 prefixes. At its IPv4 interface, all received packets start >>>> with one of them. At its IPv6 interface, all packets it receives have an >>>> embedded address that starts with one of these prefixes. >>>> >>> >>> It looks like that you treat default mapping rule as a basic mapping rule >>> to check source address consistency on the BR. It should work, >> >> Yes, thanks. >> >>> and the MAP too. >> >> I don't see that. >> The bit that, in 4rd-U hub&spoke, is changed for the CE-to-BR direction has >> been included precisely to support this case. >> Otherwise, packets sent by a CE to another CE would be routed directly to >> its destination, without a detour via a BR. >> > > I don't think so. I just mention BR behavior here.
Yes, but AFAIK the solution needs to involve CEs > You mention CE behavior for whether the bit is set or not. Yes, please see below. > >> >>>>> I think that operators who already deploy such dual-stack network is >>>>> supposed that they have address mapping table, >>>> >>>> I would rather suppose that ISPs that have added IPv6-prefix delegation, >>>> say /56s, to an existing IPv4 network did it without mixing their IPv6 >>>> plan with their IPv4 prefixes. >>>> I am ready, however, to look seriously at individual cases where choices >>>> were different. >>> >>> Basically provision MAP CE is based on its delegated IPv6 prefix in >>> concept. It is opposed to your case but technically possible. >> >> >>> Now I concern that it requires much complicated CE implementation. >> >> All what is required is that CEs set an address bit if hub&spoke topology is >> required. >> > > So how CE decide to set the bit, and when the CE figure it out? The CE knows it must set this bit if, and only if, it received at initialization a Topology-variant parameter set to Hub&spoke (sec 4.1). In this case, the CE sets bit 79 to 1 in IPv6 destination addresses of all packets it sends. BTW, this bit should better, for clarity, be given a name, e.g. bit B meaning To-BR bit (or whatever better idea one could propose). I plan to do it in the next version. RD > --satoru > > _______________________________________________ Softwires mailing list [email protected] https://www.ietf.org/mailman/listinfo/softwires
