Hi Scott,
On 2009-02-05 02:01, Scott Brim wrote:
> Excerpts from Brian E Carpenter on Wed, Feb 04, 2009 09:18:52AM +1300:
>>> Let's say the site was using 10/8
>>>
>>> The xTR in Europe would advertise 10/8 into the ALT. The xTR in
>>> NZ would advertise 10/8 into the ALT
>>>
>>> If you're in NZ, your map-request would be routed over the ALT to
>>> that xTR If you're in Europe, your map-request would be routed
>>> over the ALT to that xTR.
>>>
>>> If it is desired, the EID owner can modify the RLOC advertisements
>>> so that encapsulated data packets will go to the RLOC he wants
>>> them to.
>>>
>>> If you want to move the EID-prefix to a different geographical
>>> location, (NZ to Timbuctu) withdraw the route from NZ and start
>>> advertising it from Timbuctu. Yes the RLOC will change. The ALT
>>> hierarchy doesn't change.
>> That sounds like policy-based mapping and would create interesting
>> commercial side-effects just like BGP policy does. It also suggests
>> that explicit scoping of map entries would be desirable, filtering
>> rules for propagation of map entries too, and maybe there would be a
>> risk of black holes in the map. That all sounds very familiar.
>
> First help me understand your terms. "Policy-based mapping" means to
> me that the result of a mapping lookup would vary depending on who was
> asking.
Yes, that is the effect, whatever the mechanism.
>
> This is different from BGP in that mappings are not propagated
> hop-by-hop. They are sent directly from replier to requestor.
> Different answers would not collide in the middle, nor need to be
> generated in the middle. There is no middle.
>
> Does "scoping of map entries" mean the same thing? Does it mean that
> different authorities would have different answers for all queriers?
I guess the answer to that depends on the specific mapping mechanism.
ALT is only one mechanism, after all. I could imagine mechanisms where
there are multiple possible sources for the map entry for a specific EID,
and that would allow for explicit scoping of a given version of the map
entry. If there can only be one authoritative source per EID, then
the notion of scope is a bit artificial. (A scope would be defined as
a set of queriers who all get the same answer, fwiw.)
>
> Also, I don't see why there would be black holes in the map unless the
> originator of mapping info put them there intentionally, or are you
> suggesting there could be more chance for configuration problems?
> There shouldn't be any non-deterministic interactions in the middle of
> the network as with a hop-by-hop dissemination protocol.
Once you allow for any kind of policy, there could be black holes
I think. In terms of my immediate previous comment, the black hole
scope would be the set of queriers who all get no answer. That could
be intentional or accidental.
> Could you please be more explicit about "this all sounds familiar"?
I mean that once you have policy mechanisms in the map, you will
get the same sort of interaction between the policy mechanisms and
business relationships or AUPs that we have in BGP. Who knows where
that might lead?
Brian
_______________________________________________
rrg mailing list
[email protected]
http://www.irtf.org/mailman/listinfo/rrg
