Short version: I support Dimitri's suggested changes about the
roles of Locators and Identifiers.
The current draft still contains an erroneous
inference that the one address could be used as
both a locator (RLOC) and an identifier (EID).
The draft should be improved to better explain
the purpose of EID addresses and to show that
LISP involves separate roles for EIDs and RLOCs
- not separate namespaces, as with HIP.
Links to all versions of the draft LISP WG Charter.
Hi Dimitri,
I think your suggested improvement to the draft LISP WG Charter is
good:
>> Locator/Identifier separation approaches are rooted in the premise
>> that the current Internet architecture often uses a single IP
>> address
>> (i.e. name) for two distinctly different roles: Routing Locators
>> (which describe "where" you are attached to the network) and
>> Identifiers (which describe "who" you are)
>
> in order to prevent confusion between a name and an address, it would be
> preferable to state:
>
> Locator/Identifier separation approaches are rooted in the premise that
> the current Internet architecture designates by a single number space,
> i.e., the IP address, two distinctly different roles: Routing Locators
> (which describe "where" you are attached to the network) and Identifiers
> (which describe "who" you are).
I stopped trying to suggest improvements on the LISP list after my
concerted effort last week - 15 messages and a research project
into the meaning of "namespace" - was ignored. Links to these
messages are here:
http://www.ietf.org/mail-archive/web/lisp/current/msg00371.html
Sam Hartman suggested there may be multiple meanings of "namespace",
but I could find only one:
http://www.firstpr.com.au/ip/ivip/namespace/
My efforts included trying to point out that "Typically, the same IP
address will not be used as an identifier and locator in LISP." is
misleading, since it indicates it may be allowable to do something
which is prohibited by all 13 versions of the central LISP I-D:
http://tools.ietf.org/html/draft-farinacci-lisp-12#page-8
"EIDs MUST NOT be used as LISP RLOCs."
No-one has shown where using the same IP address as an EID and an
RLOC is allowed for in the LISP I-Ds or how it could work in
practice. I have argued in detail it cannot work.
My suggested changes are not criticisms of LISP - just an attempt at
making the Charter a more accurate and helpful explanation.
My suggested changes included better general explanation of how LISP
helps with routing scalability and that EID addresses are intended
for end-user networks which need multihoming and/or portability of
their address space. I suggested including the term "core-edge
separation scheme" as a better overall description for the class of
potentially practical solutions which LISP is one of. This is the
best term for the most promising class of scalable routing
solutions, following this paper:
Towards a Future Internet Architecture: Arguments for
Separating Edges from Transit Core
Dan Jen, Lixia Zhang, Lan Wang, Beichuan Zhang
http://conferences.sigcomm.org/hotnets/2008/papers/18.pdf
I also suggested changes which made it clear that (contrary to
repeated statements in presentations, and inferences in I-Ds) LISP
does not involve separate namespaces for EIDs and RLOCs. HIP does,
and this makes HIP impossible to introduce incrementally and
voluntarily - all participating hosts need new (IPv6) stacks with
HIP extensions and perhaps (I am not sure) HIP-compliant
applications. A scalable routing solution needs to be voluntarily
adopted by many or most multihoming-needing end-user networks.
This will only occur if the scheme provides immediate benefits,
including handling packets sent from non-upgraded networks, and if
it requires no host changes.
The big deal about LISP, APT, Ivip, TRRP and Six/One Router is that
these core-edge separation solutions to the routing scaling problem
do not involve separate namespaces - or as you describe it, in
accordance with the draft Charter "number spaces".
As your text makes clear, the nature of the separation is in terms
of "roles". HIP is arguably the architecture which truly separates
Locators and Identifiers. I think this is important to state in the
WG Charter, since LISP's name "Locator Identifier Separation
Protocol" seems to imply it is the primary such protocol.
In LISP, a subset (in multiple separate prefixes) of the global
unicast addresses which are reserved for use as EIDs - addresses for
hosts in end-user networks with LISP-mapped prefixes. The remainder
of the global unicast addresses are can be used as RLOCs: for ITR
and ETR addresses.
(However, in general LISP usage, the term "RLOC" is used more
broadly than the above description, which is based on the
formal definition of terms in draft-farinacci-lisp. "RLOC"
is also frequently used to refer to any non-EID address, such
as those of hosts and routers which are not involved in LISP.)
The different way of interpreting EID addresses affects the behaviour
of ITRs and ETRs. Hosts and all routers other than ITRs and ETRs have
no knowledge of these differing roles and continue to interpret all
these addresses as they do today - according to how they are
interpreted as part of the global unicast namespace.
My suggested changes were mainly to the first half of paragraph 3 of
the initial version - which has been expanded to paragraphs 3 to 4 in
the latest version.
Your suggested changes are to paragraph 2 and I think they compatible
with what I suggested.
However, here is a tweak of your last two lines, to remove the "you":
i.e., the IP address, two distinctly different roles: Routing
Locators (which enable routers to forward a packet to the correct
destination network) and Identifiers (which uniquely identify a
stack on an interface on a host in an LISP-enabled end-user
network).
The brief definitions I gave above strike me as apt for LISP. More
comprehensive definitions are possible, but there seems to be a
strong aversion to anything which is not terse.
The IRTF Routing Research Group is currently debating the definitions
of "Locator", "Identifier" and "Address". I think it is pointless
trying to agree on precise definitions for these terms, because in
practice people will use them in different ways depending on the
particular conceptual framework they are using.
Below are links to the various drafts of the Charter and to the more
extensive of the suggested changes. The version numbers are mine.
- Robin
Draft 00 (my numbering scheme) from Jari:
http://www.ietf.org/mail-archive/web/lisp/current/msg00262.html March 13
My suggested changes to 00:
http://www.ietf.org/mail-archive/web/lisp/current/msg00286.html March 19
Sam's versions 01 and 02:
http://www.ietf.org/mail-archive/web/lisp/current/msg00314.html March 23
http://www.ietf.org/mail-archive/web/lisp/current/msg00342.html March 25
Dow's suggested changes to 02:
http://www.ietf.org/mail-archive/web/lisp/current/msg00355.html March 25
Version 03 presented to the IETF and IESG lists:
http://www.ietf.org/mail-archive/web/lisp/current/msg00380.html March 29
http://www.ietf.org/mail-archive/web/ietf/current/msg56361.html
_______________________________________________
Ietf@xxxxxxxx
https://www.ietf.org/mailman/listinfo/ietf