Robin,

Please see below for some follow-up: 

> -----Original Message-----
> From: Robin Whittle [mailto:[email protected]] 
> Sent: Monday, February 02, 2009 8:36 AM
> To: RRG
> Cc: Templin, Fred L
> Subject: Re: [rrg] RANGER-06
> 
> Hi Fred,
> 
> Thanks for these clarifications.  I am surprised, based on what I
> read in the I that RANGER I-D:
> 
>   http://tools.ietf.org/html/draft-templin-ranger-06
> 
> that RANGER resembles LISP, APT or Ivip in that it has xTRs (ITRs and
> ETRs) and a mapping system.  There is no mention of xTRs in the I-D.

Using the LISP terminology, the RANGER "Border Router (BR)"
in fact performs the xTR function. But, have we settled on
a common terminology for RRG (and perhaps other domains
also) such that all map/encaps proposals have to use that
terminology?  

> I hope you will explain this more clearly, on the RRG list and/or in
> a revised I-D - including by giving detailed examples and
> explanations of many things, such as:
> 
>  0 - The major architectural differences between RANGER and other
>      comparable schemes, such as LISP-ALT - and what benefits and
>      problems arise from RANGER compared to some alternatives.

RANGER treats each enterprise as a link and views all
BRs (i.e., all xTRs) as neighbors on the link. That
means that ordinary link-scoped neighbor discovery
mechanisms (RS/RA/NS/NA/Redirect) can be used between
BRs as if they were on-link neighbors.

RANGER is also about discovery of BRs that can be used
as exit gateways for forwarding packets to destinations
outside of the enterprise. A special subset of these
BRs are known as "Border Gateways (BGs)" and provide
a "default" routing capability. In APT, these are
known as "default mappers". In ISATAP and VET, they
are known as "Potential Router List (PRL)" routers.  

>  1 - How the mapping system works from the point of view of the
>      ITR.  This includes what actual data is in the mapping for
>      each EID or whatever it is which is a unit of address space
>      ("inner" address space, I think, in your terminology) which
>      has a common treatment at ITRs.

How the mapping system works and the actual data in the
mapping system  is covered in detail in (VET, section 5.3.3).

About EID, again this comes down to a terminology question.
I still think the LISP EID terminology is misleading in that
the thing EID refers to is an IP address that gets assigned
to an interface just like any other IP address. I'm not going
to beat on this further, but at the same time I don't want to
propagate the use of the term in RANGER if I don't have to.

>  2 - Who controls the mapping - from an administrative point of view
>      - and in a technical sense: what are the query servers which
>      respond to mapping queries.   I gather it is a global query
>      server system as with LISP-ALT or TRRP and that ITRs are all
>      caching ITRs.

There is a separate mapping service for each enterprise.
Border Gateways (which are also known as PRL routers in
ISATAP and VET, and are also known as "default mappers"
in APT) have synchronized mapping tables that cover all
inner IP prefixes reachable from within the enterprise.
Border Gateways also provide default forwarding services
for reaching inner IP prefixes that can only be reached
outside of the enterprise.

All BRs (i.e., all ITRs) cache the mappings they receive
in their Forwarding Information Base.

>      Are the mapping queries and the query servers which send back
>      the reply all operating on the "outer" address space, the
>      "inner" or both?

Using the default mapping option, Border Gateways are
themselves xTRs that receive encapsulated packets,
decapsulate them, and send encapsulated ICMP redirects
back. So, both outer and inner space are used

Using the DNS resolution option, the BR performing the
query can locate a DNS server using the outer IP address
space only and get back an outer IP (i.e., unencapsulated
DNS response). Alternatively, it can use encapsulated
DNS queries using both inner and outer address space.
This may in fact be an important consideration for
encapsulations that use IPsec. 

>  3 - How ITRs do reachability testing to ETRs, and make decisions
>      about which ETR to use.  Ivip doesn't do it this way, but LISP
>      and APT do - so as far as I know, so does RANGER.

With RANGER, mapping resolution returns a list of ETR
RLOCs that are potential next-hops toward an EID prefix.
Locator liveness testing is done in the data plane, and
if an RLOC appears to be failing, the ITR tries another.

>  4 - One or two concrete examples of site multihoming.

OK.
 
>  5 - Likewise RANGER's "site mobility" and an explanation of how this
>      differs in principle and practice from "end system mobility".

OK.

>  6 - What "reuse of outer address space" means, with some examples.
>      I gather that, for instance, IPv4 space would be "outer" in the
>      way you are depicting RANGER - analogous to LISP's RLOC space.
>      "Inner space" I think is analogous to LISP's EID space - and
>      in your preferred example is IPv6 space.

For example, the IPv4 address space behind most home
user's NAT boxes is taken from 192.168/16. That means
that that (outer) IP prefix space is spatially reused
among all such disjoint home networks.

>  7 - I understand a "legacy" host, such as an IPv4 host would keep
>      working normally, and wouldn't be able to access anything in the
>      inner address space (IPv6).  But what about existing IPv6
>      hosts in networks which don't implement RANGER?  Will they
>      still be able to communicate with any IPv6 host in any of the
>      RANGER equipped networks?

An IPv6 host on a native IPv6 link would not notice any
change if the IPv6 router were to begin using RANGER.
 
>  8 - I understand, in your preferred example, that IPv4 acts like
>      RLOC space and IPv6 like EID space.  I guess that means that
>      each network somehow only needs to have a routing system which
>      handles a subset of the IPv6 space, with the rest accessible
>      via ITRs and the mapping system.

Correct.
 
>      If you could explain clearly how RANGER achieves a reduction in
>      DFZ size, while providing multihoming, TE and portability, that
>      would be great.

Using the IPv6 example above, IPv6 prefixes go in the
mapping table of a parent enterprise only; they do not
go in the RIB of all parent enterprise border routers.

>      I guess this means the IPv4 DFZ, where
>      all the VET tunnels occur.

In the Internet core, yes - the IPv4 DFZ. In "child"
enterprises-within-enterprises, the VET tunnels occur
within the borders of those child enterprises.

>      Is there anything like an IPv6 DFZ
>      in RANGER?

Not that I can think of at the moment.

> It is always difficult explaining something new to people you you
> don't know, via a unidirectional link such as a written document.  It
> is hard to avoid using terms which mean something detailed and
> specific to the author, but mean little - or something different - to
> the reader.

Is there an agreed common terminology for RRG map/encaps?
If not, isn't it sufficient for RANGER to define its own
terms within its own scope?

> Your response has definitely helped me understand what you are trying
> to achieve, but I probably won't be diving into VET or SEAL soon.  My
> overall experience is that you describe things in such high-level
> terms, quite a few of which I don't understand in the specific way
> you do, and that I have a very hard time constructing a nuts and
> bolts model in my head of what RANGER would be like.
> 
> It helps to know the goals of RANGER - as I think you have described
> more clearly in this reply than I in the I-D.  However, in order for
> me to I feel I understand something I need to imagine a number of
> building blocks which I understand very clearly, and then imagine
> them strung together and interacting.

If you can grasp the RANGER Border Router/Gateway
building blocks, you are most of the way there. 

> Then I can start to consider whether I think the whole thing 
> will work!

Again, VET and SEAL give the details on how it works
and a fair amount it is already working in prototype
and production code.

> I probably won't have time to build such an understanding of RANGER
> in the foreseeable future, but I am keen to read anything more you
> can say about it on the list, and I am sure other people would be
> interested too.

Hope this has helped.

Fred
[email protected]

> 
>   - Robin
> 
_______________________________________________
rrg mailing list
[email protected]
http://www.irtf.org/mailman/listinfo/rrg

Reply via email to