Michel:

In response to your later request for feedback on this proposal, here are
my initial reactions:


On Wed, 27 Nov 2002, Michel Py wrote:

> Subject: GUSL proposal (very crude)
> 
> [Note: this is independent of "GUPI"]
>
Agreed. GUPI is a separate topic which requires a workable solution. 
 
> 
> GUSL
> 
> Globally Unique Site Local
> 
> 
> Goals:
> 1. Provide an allocation method of site-local addresses
>    within FEC0::/10 in order to avoid ambiguity of such
>    addresses.
> 2. Enforce the non-routability of site-local addresses
>    on the public Internet.
> 3. Clarify the use of site-local addresses for
>    inter-site traffic.
> 
These seem laudable goals, and fall within what I believe to have been the
bounds of the consensus at Atlanta.

> 
> 1. Allocation method:
> 
>    1.1 Rationale.
>        There is a need for three types of allocation:
>        - Free, automated configuration, no registration,
>          no external connection, almost unique.
>        - Free, manual or semi-automatic configuration,
>          no registration, Internet connection necessary
>          for semi-automatic configuration, unique.
>        - Fee-based, manual registration, unique.
>          Additonal properties TBD.
> 
The three allocation types may be broader in scope that the _minimum_
requirements, but, to quote Mae West: "too much of a good thing can be
wonderful."  ;-)  I see no objection to the allocation types as enumerated
above.

>    1.2 The site-local address space (FEC0::/10) will be
>        divided in 3 parts:
> 
>        1.2.1 Free, random/hash allocation, for unattended/
>              automated setups.
>              See Paul Francis / Pekka Savola
>              FEC0::/11
> 
>        1.2.2 Unregistered, free, unique, sequentially
>              allocated.
>              See Charlie Perkins.
>              FEE0::/12
> 
>        1.2.3 Registered, probably not free, geographical or
>              other allocation method, TBD.
>              FEF0::/12
> 
>    1.3 Choice of allocation method:
>  [...]
This section seems entirely reasonable to me, at least in concept.  The
details can be settled as the work progresses.

In the later note, a specification was included which seems to require
that each allocation bear a prefix length of /48.  Since we are already
dealing with a /10 block for the entire SL space, this yields a maximum of
2**38 allocatable prefixes.  I am impelled to wonder aloud: is this a
sufficient count of possible allocations to meet the anticipated demand
over the expected life of the IPv6 protocol?  We thought, at least in the
days of our innocence, that the IPv4 address space would be amply
sufficient to the need; and we have for the past ten years or so reaped
the bitter harvest of our myopia. Would it perhaps be wiser to make the
allocations on a longer default prefix length, or, alternatively, to
allocate on varying prefix lengths as the end-user's network requirements
indicate?  I suspect, based on the experience gained in working with
clients, that a great many end-user networks will rush to acquire GUSL
space, and very few of those networks are likely to _require_ an 80-bit
address block to satisfy their operational requirements.  Having stood ten
years in a corner of the IPv4 implementation space while waiting for paint
to dry, I would be greatly distressed to see like conditions arise
pertaining to IPv6 within my lifetime.
   

> 
> 2. Enforcement of global non-routability:
> 
>    2.1 Rationale.
>        Ambiguity provided some fail-safe for route leaks.
>        By removing ambiguity, we must provide additional
>        Enforcement of non-routability.
> 
>    2.2 Routers MUST have a default blackhole for FEC0::/10.
>        See Bob Hinden.
>        This blackhole MUST NOT be easily removable, as it
>        does not prevent the site from using more specific
>        prefixes within FEC0::/10
> 
This proscription concerning routing of the site-local block seems both
necessary and desirable.  


>    2.3 Routers MUST discard by default any BGP routes
>        matching FECO::/10 ge 10. See Michel Py.
>        Accepting such routes MUST require specific permit
>        statements.

Here again, the requirement seems necessary and desirable.  Should we
consider making the routing protocol requirement more general by
substituting the specfication "any exterior gateway protocol" for the very
specific "BGP"?

> 
> 3. Multiple sites:
> 
>    GUSL addresses SHOULD NOT be used for communication with
>    other sites.
>    (I am open to a MUST NOT, whatever the WG consensus is)
> 
It seems to me that item 3 above may be interpreted variously to mean,
among other things:

a) GUSL addressing may not be assigned to links (private or otherwise)
connecting two or more sites, even if the GUSL addresses are not routable
over the public network; 

or

b) traffic bearing GUSL source or destination addresses may not be carried
over links (private or otherwise) interconnecting two or more sites, even
if the links themselves bear global addresses

Based on previous traffic in the mailing list, I would expect that your
intent would have been closer to the latter.  

I see potential implementation inconveniences with either interpretation
above, but nothing that is insurmountable if a reasonable PI address
allocation scheme for private network facilities is worked out before IPv6
begins to see widespread service in end-user networks.

That's my $.02 worth.

AEB

-- 
Alan E. Beard
<[EMAIL PROTECTED]>
AEBeard Consulting
4109 Chelsea Ln
Lakeland FL 33809
863.815.2529


--------------------------------------------------------------------
IETF IPng Working Group Mailing List
IPng Home Page:                      http://playground.sun.com/ipng
FTP archive:                      ftp://playground.sun.com/pub/ipng
Direct all administrative requests to [EMAIL PROTECTED]
--------------------------------------------------------------------

Reply via email to