<<< Chronological >>> Author Index    Subject Index <<< Threads >>>

Re: FWD: [GLOBAL-V6] New draft available: IPv6 Address Allocationand Assignment Global Policy

  • To:
  • From: Havard Eidnes < >
  • Date: Fri, 26 Apr 2002 14:03:28 +0200 (CEST)

Hi,

I have a concrete question and a general comment with regards to this.

The concrete question is illustrated by an example:

 o Assume I'm a transit service provider with my own AS
 o Assume that I only sell "wholesale" service to a smaller number of
   customers
 o Assume that I want to provide IPv6 service
 o Assume that all my customers have their own IPv6 address allocations

Now, where does that leave me in terms of getting IPv6 addresses
assigned to number my internal network and the few internal servers we
have?  The way I read the draft policy, the answer would be "out in the
cold".

One possible way out would be to get an e.g. /48 assignment from one
of the downstreams, which address-space-wise would be sufficient to
number this network.  However, there is no guarantee that the route
for the /48 would not be filtered away "all over the place" (I would
guess that it *would* be filtered away).  The result would be that if
the connection to the downstream which announces the enclosing /32
goes away, so does the general connectivity to the systems in my own
AS.

Another option could be to get an address assignment from one of my
upstreams.  However, again, my general connectivity would be married
with that provider's connectivity (do to assumed filtering of /48
routes), so does not prove to be useful in a setup with more than one
"upstream" provider; if the connectivity with that provider was
broken, so would my general connectivity be, even though I have
physical connectivity via my other upstream(s).


My baggage is from IPv4, so I may have missed a few details (corrections
appreciated), but it seems to me that the attempt of imposing a strict
hierarchical address allocation covering multiple routing domains can at
best be characterized as an attempt at putting a band-aid across the
gaping chest-wound called "IPv6 multihoming", by in essence telling
people "don't do that!" and also "you do not have routing-wise autonomy
(or visibility) even though you have an AS".

The draft policy says that "routability" is not guaranteed for any
assignment or allocation, and the policies as to what is commonly
filtered and what isn't are not yet defined (I think), but how many
think /48s will be universally accepted?  (I certainly don't.)


Regards,

- Håvard




  • Post To The List:
<<< Chronological >>> Author    Subject <<< Threads >>>