On Tue, 9 Mar 2004, Mattias Pettersson wrote:
> Hi, > > [This issue spans some WGs but it originates from here I believe.] > > We've had a small discussion in NEMO about multiple default routers on a > link, and whether these DRs are allowed to advertise different prefix > sets or not. > > Think of this small network: > > > ISP A ISP B > \ / > Ra Rb > | | > -+-----+-----+- > | > H > > Further assume that Ra advertises prefix Pa and Rb advertises prefix Pb. > Host H will have two DRs, hear two prefixes and build addresses out of > the both. > > What I'm asking for is whether this scenario was ever considered > "broken" unless there is some form of coordination between Ra and Rb. I > seem to recall that there was a discussion on ipng/IPv6 on exactly this > long ago but on the other hand I can't find it. > > The reason for calling this scenario broken is that H must be careful > with what source address to use when sending to either default router > (and that is a requirement we can't put on a legacy IPv6 host). One can > assume that both ISPs perform ingress filtering. If H contructs a packet > with a source address Sa from Pa and: > > - sends it to Ra: > - the packet will be routed successfully > > - but if it was to send it to Rb instead: > - (1) ISP B would drop it due to ingress filtering, or > - (2) Rb would have to forward it to Ra, or > - (3) Rb would have to tunnel it to ISP A, or > - (4) Rb would have to send a redirect to H, or > - (5) Rb or ISP B would generate an ICMP error about bad source address > > Alternatives 2-5 are typically described in draft-huitema-multi6-hosts, > but they either: > - require some form of coordination between the routers and their > ISPs, or they > - require that the host H implement some additional mapping between > prefixes/source addresses and default routers. > > To H there is no difference between Ra and Rb. Both claim that they are > default routers. Now both should be able to forward any traffic from H. > But unless Ra and Rb are coordinated in some of the ways described above > or that host H implement _new_ requirements in > draft-huitema-multi6-hosts (which standard IPv6 hosts do not), then host > H will have to take a chance on what default router to use (last heard? > first heard?). If H happened to select Ra as default router, packets > with source Sa will go through, but no packets with source Sb. And > default router selection will not happen again until Ra becomes unreachable. > > So the question is: am I correct to regard this scenario as broken and > say it should not be encouraged? I also think that nobody plans to build > fixed networks like this, just for these reasons. However, people may > want to build NEMO-type of networks just like this, as it is convenient > when mobile routers are rather independent, they come and go, and each > connects to one ISP and each only advertises a prefix from its ISP. > Awkwardly, then it is more important than ever that the mobile routers > are cooperating otherwise legacy IPv6 hosts won't be able to get plain > connectivity. I think this is not broken at all. The host should select the correct prefix according to the source address selection rules. I tried this scenario approximately 3 years ago on a KAME stack and it was working correctly. Best Regards, Janos Mohacsi -------------------------------------------------------------------- IETF IPv6 working group mailing list [EMAIL PROTECTED] Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------