Válasz: [GLOBAL-V6] IPv6 Network Plans (Forcasting) (fwd)

[turchanyi.geza@ln.matav.hu]

Dave, Takashi,

as always, you are rising important questions.

1, The first thing to think about: when /48 should be allocated.

The IETF recommendation is talking about a /48 for each site. This
recommendation is well aligned with the 6to4 tunelling protocol, this protocol
can make distinctions between /48s at most, and smaller allocation does not
realy save address space where the real bottleneck exist today: in IPv4 space.

The next question: what is a site. Site should have at least one egress router.
A typical dial-up, DSL or Cable-modem user has no router at all, just one host.

Therefore my suggestion would be: a /48 could be allocated to those customers
having an egress router, and it is not important at all,  what is the access
method in use.

If this will create a boom at the router market, then we could gain some money
in the stock markets ;-)

2, A LIR without any customer could allocate its first /48 to its first customer
and what to do next?

In IPv4 we (at least in RIPE region) had a slow start mechanism implemented. I
would not object a somehow guided mechanism for unexperienced address
allocators, even in IPv6. A new LIR might have already experienced personel,
needing just big enough address range, and no guidence.

3, I do not think that we will be able to allocate to every potential LIRS such
a big address space that no further allocation will be necessery.  After a while
we will realise that some ISP-s have many prefixes, and others have big address
space unassigned. This could be even the case as we would be able to allocate
"infinite" range of addresses to ISPs, because ISPs can merge together.

I do not see any other mechanism to cope with this fragmentation of address
space than somehow forced and supported renumbering. We should find a way to
create continous larger block for the big ISPs, and a way to swip out old
reservations from the middle of the larger blocks, in order to enable this
larger blocks for reallocation.

My guess is that the simplest method to do is: expiration of address space, new
block allocation before the old segment would expire, end similer mechanism for
address assignements for customers. Of course, an ISP should have its allocated
address space longer than this ISP would assign address space for its customers

My initial estimation for expiration of an ISP (LIR) space would be 20+ years,
and assignement for a site can be limited at  8-10 years. The site would be able
to ask for a second assignement long before the first expires, and could have
the two assignement parallel for one year at least, to help smooth transition.

This concept is little bit more than lease of address space, I think. This
concept would allow as smooth change in the allocation policy as well, if the
allocation policy should be reviewed after a while.

Kindest regards,
               Geza Turchanyi





Dave Pratt <djp-ripe-lists@djp.net> dátum: 2001.10.26 12:54:33

Címzett:       global-v6@lists.apnic.net
Másolat:        (Vakmásolat: Turchányi Géza/PKI/HTC2)
Tárgy:         [GLOBAL-V6] IPv6 Network Plans (Forcasting) (fwd)



Hiya Folks,

I just wanted to say hello to you all, and repost something I sent to the RIPE
lists a week or so back (before this list went live).

Apologies if you have already seen this at RIPE.

Cheers
Dave

---------- Forwarded message ----------
Date: Wed, 17 Oct 2001 12:02:19 +0200 (CEST)
From: Dave Pratt <djp-ripe-lists@djp.net>
To: lir-wg@ripe.net, ipv6-wg@ripe.net
Subject: IPv6 Network Plans (Forcasting)

Hiya Folks,

I would like to elaborate on a comment made at the end of the LIR/IPv6 session
in Prague about the ability of IPv6 address space requesters to forecast their
address requirements. These suggestions probably might not make it into the
provisional policy, but should be considered for later policy.

A large ISP, probably global, and certainly national, could build its entire
infrastructure with a single /48 network block. For example this would allow
256 "locations", each with 256 networks, each with more hosts than ants on
this planet.... With the popular /127 for point-point connections even
customer lines can be accomodated.

Accordingly, the only driver for longer prefixes would seem to be the /48
customer networks. How will applicants for IPv6 address allocations quantify
these networks ?
The only way I can see is from very early marketing forcasts.

In the past RIPE have viewed marketing forecasts with a lot of suspicion
(quite rightly). However, unless RIPE issue a "one size fits all" prefix they
may have to base allocations on these marketing forecasts.

One thing however can be relatively easily determined, and that is "whether an
applicant has ANY customers".

For those LIR WITHOUT any customers, we could allocate a /48 and even large
multinational corporations would have plenty of addresses (see second
paragraph above again).

For those LIR WITH customers, we should be much more generous because we
cannot accurately judge needs from marketing forecasts. Assuming a HD ratio of
0.8, then a moderately successful city carrier LIR with 100,000 DSL or Cable
customers (each requiring a /48) will qualify for more than a /28 allocation.

Do we want such LIRs to come back and ask for a second route (in the global
routing table) ?

Conclusion:
1. We should consider allocating a /48 to LIR without any customers as this
is plenty.
2. Even a /28 would result in considerable numbers of followup requests if
cable/DSL customers (mass users) get /48.

Opinions, Comments, Concensus ?

Cheers
Dave






-
- This list (global-v6) is handled by majordomo@lists.apnic.net