RE: draft-pioxfolks-6man-pio-exclusive-bit-02.txt

["Templin, Fred L" <[e-mail]>]

Hi Bob,

> -----Original Message-----
> From: ipv6 [mailto:[e-mail]] On Behalf Of Bob Hinden
> Sent: Thursday, January 25, 2018 7:52 AM
> To: Ole Trøan <[e-mail]>
> Cc: james woodyatt <[e-mail]>; Bob Hinden <[e-mail]>; IPv6 List <[e-mail]>
> Subject: Re: draft-pioxfolks-6man-pio-exclusive-bit-02.txt
> 
> Ole,
> 
> > On Jan 25, 2018, at 12:22 AM, Ole Troan <[e-mail]> wrote:
> >
> > Brian,
> >
> >>>> But "L2 isolation is in effect" and "you are the only host in this
> >>>> prefix" are orthogonal; one, both or neither could be true. So
> >>>> two bits are needed.
> >>>
> >>>
> >>> They are not orthogonal.
> >>
> >> Well, that's a matter of definition. Neither one implies the state of the
> >> other, which is what I meant by "orthogonal".
> >>
> >>> They are entangled because the latter is not easy to do unless the former is done as well.
> >>>
> >>> Consider the specification complexity of each of the following three possible combinations.
> >>>
> >>> * Shared link + Shared prefix: status quo, already specified.
> >>>
> >>> * Isolated link + Shared prefix: very nearly status quo; host MAY safely ignore all ND Redirect, which is pretty much a trivial
> refinement of what hosts are already doing today. They still MUST perform DAD and address defense.
> >>>
> >>> * Isolated link + Exclusive prefix: not too different from Isolated link + Shared prefix, above, except hosts could also dispense with
> DAD and address defense.
> >>>
> >>> Unfortunately, the expired PIO-X draft presents the remaining combination:
> >>>
> >>> * Shared link + Exclusive prefix: we need to define rules for how hosts behave in the face of conflicting X-bits on matching PIO
> advertisements from multiple routers; including how to behave when routers send ND redirects that conflict with other routers PIO-X
> advertisements, we need a better way to redirect whole prefixes and not just addresses, e.g., the also expired I-D.templin-6man-rio-
> redirect. This is a huge mess compared to the other combinations.
> >>
> >> Exactly. There are four combinations, and each has to work. We agree on that.
> >
> > I don't think we agree on "each has to work".
> > We could very well define an initial version of PIO-X that is restricted to P2P links (or something that is made to appear to look like
> one).
> 
> Right.  I think it makes sense to focus on the Isolated link cases.

PIO-X currently defines an *unsolicited* mode of operation where the
router provides the host with an exclusive prefix that the host did not
ask for. There should also be a *solicited* mode where the host can
ask for an exclusive prefix, e.g., by including a PIO in its RS messages
to the router. In that case, it would begin to look like DHCPv6 PD.

> As you noted, it’s easy to create isolated links on shared media.  As
> stated the “shared-link + exclusive prefix” gets complex very fast, and overlaps with other solutions such as using routing protocols.

Very large shared links may not be candidates for using routing protocols.
Some NBMA links may connect thousands of nodes or more, for example.

> IMHO, better to focus on things that we can make work.

DHCPv6 PD falls under the scope of things that can work.

Thanks - Fred

> Bob