[ppml] "Recommended Practices" procedure

[Scott Leibrand]

On 06/29/06 at 12:20pm -0700, Owen DeLong <owen at delong.com> wrote:

> --On June 29, 2006 2:56:25 PM -0400 Scott Leibrand <sleibrand at internap.com>
> wrote:
>
> > Another consideration is that a PA /48 need not be accepted globally to be
> > usable for multihoming.  If both your transit providers accept your /48
> > from you and from each other, you can be guaranteed reachability.  (You
> > may not be able to do the kind of traffic engineering you might want,
> > though.)
> >
> If their upstreams don't accept it, then, no, you aren't guaranteed
> reachability.  You're just slightly less subject to MOST of the things
> that take out PART of one of the providers.

I'm sorry, I made an unstated assumption that you're buying transit from
two transit providers who don't have any transit providers of their own,
just peers.  IOW, tier 1 NSPs.  Can you enumerate any failure modes in
that case, or were you just talking about reachability problems for tier 2
NSPs?

> I think the cooperating filter policy suggestion is about the best way to
> handle this.  If two ISPs want to cooperate and open holes in their PA
> blocks with each other, that doesn't mean anyone else has to.  Yes, it
> does mean multihoming for those customers is slightly less reliable than
> for customers using PI space, but, I don't see a big downside to that.

I agree.  However, I would recommend suggesting it in a "Recommended
Practices" document, as doing so doesn't pollute the global table, just
yours and your peers'.

-Scott