Dave Katz <[email protected]> wrote on 07/08/2009 00:04:59:
>
>
> On Aug 6, 2009, at 1:28 PM, Joakim Tjernlund wrote:
>
> > Dave Katz <[email protected]> wrote on 06/08/2009 20:24:25:
> >
> > Hi Dave :)
> >
> >>
> >> If there is no back link, there is no link, and SPF moves on.  If the
> >> network is partitioned, somebody should fix it.
> >
> > Yes, this problem exist for a short while, until the remote
> > route has updated it's LSA.
> >
> >>
> >> Ignoring the bidirectional test rule can lead to loops and black
> >> holes, particularly if other implementations are following the rules.
> >
> > hmm, can you be more specfic?
> >
> > Consider this example:
> >
> > R1 R2 R3
> > |  |  |
> > ------- N1
> >
> > Destination is R3 and R2 is the calculation router. R2 cannot find a
> > backlink
> > so it falls back to the intervening router case and uses the nexthops
> > from the "parent", hoping that one of them will redirect IP frames to
> > R3.
>
> This is an uninteresting case, as there are no alternative paths.

Yes, it is very simple to make it clear what I am after.

> Further, there will be no "intervening router" in this case, since
> each of R1, 2, and 3 will be adjacent to only N1.

Yes, I am just saying that in this particular case I want apply the
same SPF rules as the intervening router uses.

> There will be no
> next hops to inherit.  The "intervening router" rule basically just
> points out that if the best path from A to C is through B, the next
> hop to C is simply the next hop to B.  More or less.
>
> Where it can get much uglier is if there is a back path between R2 and
> R3;  other routers playing properly will try to send the traffic
> through R3, but R3 will try to forward it locally.

hmm, don't you mean "through R2, but R2" ?

> In this case it
> will probably result in the traffic just disappearing down a hole
> instead of being properly delivered on the alternative path.  But if
> you start playing these games more generally, you can end up with
> traffic flying in circles.

I am starting to think that my idea was a bad idea now :)

>
> While routing is transiently nondeterministic, "hoping" is a bit less
> rigorous than required.  ;-)  And by "redirect" I assume you mean
> "forward", as "redirect" means something specific (and only applies to
> hosts.)

Yes, I meant forwarding(sloppy me :)

>
> In the general case, routing only converges on stability because
> everyone plays the game the same way, and the rules are designed to
> make it happen.  If you change the rules or somebody plays them
> differently, it is very easy to break the algorithm.
>
> It's not at all clear to me why this would be interesting, even if it
> worked.  In a modern, competent OSPF implementation (namely, one which
> ignores the rules about repeated LSA generation and is otherwise
> carefully tuned) you will see the other guy's LSA with the back link
> within a few milliseconds anyhow.  In a well-engineered network there
> will be an alternative path anyhow, and traffic can be delivered on
> that path until the new one comes up fully.
>
>

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

Reply via email to