On Sun, 6 Mar 2011 11:40:12 +0100 (CET) Mikael Abrahamsson <[email protected]> wrote:
> On Sun, 6 Mar 2011, Mark Smith wrote: > > > I don't think I said an on-link prefix was required. All I have ever > > said is that, as per the RFC5942, if you want to have an on-link > > prefix, you must announce it in a PIO (without the A bit if you don't > > want it to be used for SLAAC). > > You said: > > > I don't think that will always work. The PIO is needed to indicate to > > end-nodes what the onlink prefix(es) are, as per RFC5942. > > I interpreted your use of "needed" as "required". > As always context matters. I was stating it in the common and usually unstated conext of the most common case. Usually people want optimal traffic forwarding paths, which on a multi-access, peer-to-peer layer 2 segment (e.g. an Ethernet), means traffic is sent directly between attached nodes, rather than "hair-pinned" or "trumbonned" through a default router. In this common case, you'll need a PIO option with the L bit switched on, to indicate the on-link prefix(es), regardless of the addressing information and address prefix length that DHCPv6 supplies. In the less common cases of "non-link-local prefix-free" unnumbered links, you don't need a PIO. In the less common case of DHCPv6 only (M+O in RA) with hair-pinned traffic paths via a router on a multi-access, peer-to-peer layer 2 segent, you don't need a PIO. In the less common case of SLAAC only (-M -O in RA), with hair-pinned traffic paths via a router on a multi-access, peer-to-peer layer 2 segment, you'll need a PIO, with the A bit on, and the L bit switched off. > > Steinar gave an example which could imply that wasn't the case, but his > > example was actually one where an on-link prefix wasn't required. IOW, > > his example didn't disprove my statement, it only showed that there are > > some situations where no PIO announced prefix has a use case - where all > > destinations are required to be considered off-link. > > Well, I interpreted your "I don't think that will always work" as saying > that you thought what he said wouldn't work in all sitations. What > situations were you thinking of that won't work? > > As far as I can see, not having an on-link prefix means all traffic will > be passing via the router, but that's not the same as "not working", it > might just be suboptimal in some deployment scenarios and highly desired > in others. > > -- > Mikael Abrahamsson email: [email protected] -------------------------------------------------------------------- IETF IPv6 working group mailing list [email protected] Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------
