Short version: I support statements 1 and 4.
Statements 2 and 3 concern RRG consensus, which only
the co-chairs can decide.
Statement 2 also seems to assume that the difference
between CEE architectures (e.g. GSE, ILNP or
GLI-split) and CES architectures (e.g. Ivip or LISP)
is merely an "engineering" matter. I argue that the
differences are architectural and are highly
significant.
Hi Tony,
I didn't vote because questions 2 and 3 concern whether the RRG has
consensus or rough consensus. I can't tell whether there is
consensus or not - only you and Lixia can do that, by whatever
measurement techniques and thresholds you choose.
Here are my thoughts on the four questions:
(1) "The Internet continuing down the current architectural path,
whereby site multi-homing increases the size/entropy of
the DFZ RIB/FIB, is not believed to be scalable or viable."
I entirely support this statement. The current growth in the DFZ
RIB/FIB is unsustainable in the long-term. That is, I do not expect
the financial and other costs it creates to be reasonable now or
sustainable in the long-term, considering all the technological
progress we might expect in routers.
This growth is only a part of the multihoming aspect of the routing
scaling problem. The other part concerns the networks which can't
get multihoming due to the current restrictions. The same goes for
portability.
The current growth: http://bgp.potaroo.net is due to a highly
constrained version of the growth in multihoming and portable address
space: only those networks with the resources and the desire
(considering the negative impact it has on all DFZ router operators)
to gain and advertise their own PI space are in fact getting
multihoming and portability.
If there was a less expensive, less onerous and less unfairly
burdensome approach to multihoming and portability then the true
demand for these would become apparent.
Furthermore, I believe there is a demand for global mobility of IP
addresses, which would involve a vastly greater number of "networks"
- potentially every IP-connected cellphone in the world. (And I read
that India is adding 20 million cellphones a month.) The only
approach I am aware of which can achieve this is:
http://www.firstpr.com.au/ip/ivip/TTR-Mobility.pdf
(2) "The RG has rough consensus that separating identity from
location is desirable and technically feasible. However,
the RG does NOT have consensus on the best engineering
approach to such an identity/location split."
Ran provides no definition of "separating identity from location".
I believe that Locator/Identity Separation is implemented by HIP,
GSE, GLI-Split, ILNP, Name-Based Sockets and RANGI. ("Core Edge
Elimination" architectures.) With these, every host has one or more
identities and one or more locators - with Identifiers and Locators
being in different namespaces. Host protocols are changed
accordingly from the current model in which the IP address performs
both roles.
The other major class of scalable routing proposal as the Core Edge
Separation (CES) architectures: IRON-RANGER, Ivip, LISP and TIDR.
There are no host changes. Each host has one or more IP addresses
and each IP address functions as both an Identifier and a Locator (of
the host's particular interface, not of the host). So hosts do not
in any way gain separate identifiers and locators. The fact that
some new elements of the routing system look up some mapping to find
what in LISP is regarded as an RLOC (Routing Locator) doesn't mean
that there is "separation of identity from location". The RLOC, ETR
address or whatever is not seen by hosts and can applies to entire
networks, not just to individual hosts.
If the RRG discussion and consensus continues on the basis that these
CES architectures also involve "separating identity from location"
then I think this would be doing the field a great disservice, by
fudging an architecturally important distinction.
I think the statement implies that both these types of architectures
(CEE and CES) involve "locator - identity separation" and that the
differences between them all are only "engineering" differences, not
**architectural** differences.
It is fundamentally architecturally different to give each host (or
host interface) separate Identifiers and Locators than to leave them
as they are, and solve the scaling problem in the network alone (CES
architectures). This is because the former (CEE) involves
fundamental changes to the host protocols - including placing extra
burdens of computational effort, management traffic and consequent
delays on hosts. Who would seriously argue (rather than simply
assert) that there is no significant, architectural, difference
between, for instance, ILNP and LISP, or ILNP and Ivip?
For more on this, please see the threads in recent months: "LISP does
not implement Locator / Identity Separation" starting:
http://www.ietf.org/mail-archive/web/rrg/current/msg06190.html
and "Why won't supporters of Loc/ID Separation (CEE) argue their
case? (Summary of differences)":
http://www.ietf.org/mail-archive/web/rrg/current/msg06250.html
(3) "The RG has consensus that the Internet needs to support
multi-homing in a manner that scales well and does not have
prohibitive costs."
I agree that "the Internet needs to support multi-homing in a manner
that scales well and does not have prohibitive costs." Likewise
portability, mobility and multihoming with inbound traffic engineering.
(4) "Any IETF solution to Internet scaling has to not only
support multi-homing, but address the real-world constraints
of the end customers (large and small)."
I agree with this. We can't impose solutions. They have to be
adopted very widely on an entirely voluntary basis. I believe the
solution has to work and be attractive for networks of all sizes
which have any significant need for what is currently hard to get, or
unobtainable: multihoming (with inbound TE), portability and mobility.
If we make a solution which is not perceived to be suitable for the
largest networks, then many smaller networks would avoid it, because
they plan to become bigger in the future. For more on this, please see:
http://www.firstpr.com.au/ip/ivip/RRG-2009/constraints/
- Robin
_______________________________________________
rrg mailing list
[email protected]
http://www.irtf.org/mailman/listinfo/rrg
