Hi Ole,

> -----邮件原件-----
> 发件人: Ole Troan [mailto:[email protected]] 代表 Ole Troan
> 发送时间: 2010年8月2日 19:17
> 收件人: xuxiaohu 41208
> 抄送: [email protected]
> 主题: Re: [Softwires] About draft-guo-softwire-6rd-ipv6-config-00
> 
> Xiaohu,
> 
> > First, as stated in the DHCPv6 specification [RFC3315], "...The client
MUST
> use a link-local address assigned to the interface for which it is
requesting
> configuration information as the source address in the header of the IP
> datagram." Since the link-local address can not travel through the 6rd
domain,
> the CPE SHOULD act as a DHCPv6 relay agent. I think nobody will argue
against
> this conclusion.
> 
> do you only need "stateless" DHCPv6?
> if so, I think the DHCPv6 specifications should loosen up on the
restrictions
> of which source and destinations can be used. then you can just use a
global
> address as a source and the IPv6 address of the BR as the relay/server. (I
> believe this was Townsley's proposal via IM during the softwires meeting).

If the CPE could work as a DHCP relay agent which sends DHCP-relay-forward
messages with its global IPv6 address being the source address, the DHCPv6
specification doesn't need any change. 

> > Second, as I mentioned during the presentation, relaying DHCP request
> messages to All_DHCP_Servers multicast address by CPEs is not optimal in
6rd
> scenario. Note that here what I said is "not optimal", rather than "not
> possible". The CPE as a relay agent, could be upgraded to relay
> information-request DHCPv6 messages (multicast) towards the 6rd BRs.
However,
> with this approach, the DHCPv6 servers would have to be located in the
IPv6
> Interne
> > t, rather than in a 6rd site which is owned by the 6rd SP. In addition,
this
> usage will result in more traffic burden on the 6rd BRs.
> 
> I don't quite see that this has to be the case. the DHCPv6 server can
obviously
> also be part of the SP's network. it is just placed on an "IPv6 leg" of
the
> BR or the BR acts as the DHCPv6 server. the multicast packets could very
well
> be encapsulated in unicast L2 (IPv4) to the BR.

Yes, it works. However, it's not optimal. First, if a SP deployed multiple
BRs for load-balancing, each BR should be enabled as a DHCP server or a DHCP
relay agent. Second, in case that the DHCP server/relay on a given BR is
unavailable due to some reason but the other functions of it (e.g., the
anycast route for it) are still available, once the DHCP message is
encapsulated in unicast L2 (IPv4) to that BR, the DHCP service is
unavailable any more. That's to say, it's hard to achieve the high
availability for DHCP servers/relay agents.

> >
> > Hence, we propose that the CPE as a DHCP relay agent, SHOULD know at
least
> one DHCPv6 server address.
> 
> of course, the other option is to just have a DNS proxy on the CPE.

Yes, in case that DNS is the only configuration information needed.

> in any case I would like to see these requirements in the next phase of
the
> requirements for IPv6 routers in v6ops. and I'd like to start a discussion
in
> DHC about this as well as some other clarifications needed for CPE
routers.

Good idea.

Best wishes,
Xiaohu

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

Reply via email to