Re: requirements draft revision

[Joe Abley <jabley-ietf@automagic.org>]

On Wed, Jun 27, 2001 at 11:54:32AM -0700, Tony Hain wrote:
> Joe Abley wrote:
> > My point was that it is trivial to affect the routing policy of
> > an external organisation by buying transit from them.
> Then why couldn't 'popular site' trivially affect the routing policy of 'one
> particular provider' to bypass the path? After all they bought transit from
> them.

They might be able to say to that one provider "don't propogate my
advertisements to the promising regional ISP with whom you have
peering congestion". However, the result of that is quite possibly
that they will become *completely* isolated from that regional ISP,
which is not their intention.

They want their prefixes to propogate there in a manner which does
not cause traffic following them to experience congestion.

> All that a site can hope to affect is the 'internal' routing policy of a
> service provider, and then only because that is the service being bought.

Correct. That is my point.

> > If I made the first paragraph "between its transit providers"
> > rather than "between transit providers" would thae make it
> > clearer?
> It would certainly limit the scope of the statement to what is achievable
> today.

I don't actually think it limits its scope at all, but if it lowers
the chance of misunderstanding, I'm happy to put it in.

> > The requirement here is that "popular site" should be able
> > to multi-home between ISP4 and ISP5 in order to avoid
> > congestion between ISP4 and ISP5.
> Basically what you are saying here is that 'any site should be able to
> acquire services from an arbitrary set of providers'. But the statement in
> the draft goes significantly further:
> 	"The multihoming architecture should allow E to ensure that
>       in normal operation none of its traffic is carried over the
>       congested interconnection T1-T2."
> This implies that any multi-homed E has realtime control of global routing
> policies of all other organizations.

No it doesn't. It implies that E has a mechanism to influence the
way its traffic is carried within each of T1 and T2.

> The best that E can hope for is that
> traffic from direct customers of T1 & T2 not be carried over the congested
> path.

No. Traffic from beyond T1 and T2 ultimately has to be carried by T1
and T2 in order to reach E. Once in T1 or T2, the same forwarding
decisions will be made for all regardless of the source of the traffic.
If T1 never sends traffic to T2 in order to reach E, that holds for
all traffic sources.

> It may through selective announcements be able to affect some traffic
> beyond the direct customers of T1 & T2, but it can't do that reliably on a
> global scale.

The TE requirements of T1 and T2 are not relevant.

> > I still don't see what is unobtainable about the requirement,
> > given that it describes motivations for multi-homing that are
> > achievable today with v4/CIDR-abuse.
> The words used are overly broad compared to actual practice. I don't have a
> problem with requiring that current practice be continued, but make sure the
> words are not setting us up to require something beyond reality.

I don't understand how you reach some of your conclusions (above), and
am really not seeing where the "words used are overly broad".

If you would like to point out the particular phrases that are causing
you to draw some of the conclusions above, that might be useful.


Joe