On 11/11/08 3:39 PM, William Herrin allegedly wrote:
> Hi Folks,
>
> I'm trying to put together a more or less concise summary of the
> general architectures we've discussed here these past couple years.
> This is not a comparison of specific proposals (which Robin Whittle
> has done an excellent job of) but rather a summary of the universe of
> general strategies we've looked at and haven't resolutely rejected.
> I'd appreciate your constructive criticism:
>
> http://bill.herrin.us/network/rrgarchitectures.html
>
>
> Particular answers I'm interested in:
>
> 1. Have I overlooked any viable approaches to the problem? If so, what are
> they?
>
> 2. Have I overlooked any architectural elements? I'm looking for
> architectural elements here, not engineering issues. For example, I
> left out path-MTU issues because that's a "how do we shoehorn this
> into IPv4 of IPv6" engineering issue. It's only relevant in an
> engineering compatibility context. Obviously engineering compatibility
> issues will greatly inform the final architecture, but that's not what
> I'm after in this document.
>
> 3. Do you see any areas where I could offer a more clear description?
> How would you word it?
>
> 4. Have I listed anything for which we have a strong consensus that we
> can discard the approach from consideration due to some uncorrectable
> defect which is obvious even without an engineering viability study?
> By strong consensus, I mean "nearly unanimous."
Hi Bill. Here are some notes.
Since you start by talking about the root cause of the routing scaling
problem, I'll start by saying that it isn't _just_ conflation of locator
and identifier. Even if identifiers and locators were completely
separate, if routing stayed the way it is today you would still have
problems because sites will still want PI prefixes if only to avoid site
renumbering problems, and will still inject prefixes for traffic
"engineering" and to prevent hijacking.
Also there are three different identification issues that are themselves
conflated much of the time:
(1) AAA identifiers used to check in with a home agent and/or source
of funds and access a network at all.
(2) Persistent identifiers that can be used for location discovery.
They could even be DNS names, SIP URIs, HIP HIs, GSE ESIDs,
whatever.
(3) Potentially ephemeral identifiers for session control.
Some of the schemes we've been discussing separate #2 from IP addresses.
Others separate #3. Others don't separate them at all (leaving that up
to something else) but still solve routing scaling anyway.
Onward ...
It looks like Strategy A includes both what UCLA calls "separation" and
translation, and that Strategy B is "elimination" with the assumption of
rewriting at the edge. I think there are four categories here, and they
are different enough that they should be separated. In particular,
separation and elimination preserve packet headers end-to-end. OK maybe
not in the face of carrier grade NAT but at least there is a chance that
headers will be preserved. Translation rewrites headers at least once,
at the core boundary. This gives you three cases: Separation with
headers maintained (e.g. map-n-encap), Elimination with headers
maintained (e.g. ILNP), Translation (e.g. Six/One Router). Finally you
have a fourth, the GSE special hybrid which includes Elimination by
splitting ID and Locator at the endpoint, but also Translation at the
network edge.
Scott
_______________________________________________
rrg mailing list
[email protected]
https://www.irtf.org/mailman/listinfo/rrg
