[ppml] Re: [address-policy-wg] Is the time for conservation over?

[Michel Py]

Leo,

> Leo Bicknell wrote:
> Check out http://arneill-py.sacramento.ca.us/ipv6mh/.
> Note, I think the newest drafts are now called MHAP, not ipv6mh
> anymore, but the historical data is still on the other page.
> Part of the problem is that it's not all quite standard yet.

And it will never be. I have abandoned the development of MHAP for
pragmatic reasons: my assessment is that it would never be a success
because IPv6 will never be a success. I closed the MHAP list last week,
the reason being that I joined the anti-v6 camp. I will post soon a
rationale for it, still polishing the text. Basically, a simple matter
of money: know when to take a loss and move on. This hurts me, but
putting my head in the sand leaving another part of my body exposed did
not look the best thing to do.

Consolidating with your other posts:

>> David Conrad wrote:
>> The obvious implication of this is that it pretty much guarantees
>> the creation of a new and potentially much bigger swamp.

> Leo Bicknell wrote:
> A new swamp, yes. Bigger, I'm not so sure.

I share David's concerns though. The potential for a much bigger and
much murkier swamp is here. To be detailed later, one of the reasons I
am moving away from v6 is that my assessment is indeed that the IPv6
soup will indeed be worse than the IPv4 one, see at the end of this
post.


> If every ASN had a single IPv6 /32 the IPv6 table would be 12%
> of the current IPv4 routing table.

This topic has been extensively discussed on ipv6mh in the past, and
it's half-true and half-false. The issue is that "If every ASN had a
single IPv6 /32" does not register. Large multihomers need either
multiple /32s or to announce subsets (such as /36) of their own /32,
which for all practical purposes does not change much in terms of number
of entries. There is potential for much more than 12%.

> I don't think that something along the lines of "everyone with an asn
> and one or more IPv4 blocks can receive one IPv6 block of appropriate
> size" would be a bad thing.  Indeed, I'd rather see each ASN with it's
> own /32 than see some get /48's from three providers and start
>  reannouncing them (which is a big part of the current IPv4 table
> bloat).

Pekka Savola has proposed something similar a year ago, it did not move
ahead on the grounds that it would trigger a land rush on ASNs and
precipitate the adoption of 32-bit ASNs for no real gain.

Note that in terms of what is being announced in the global IPv6 routing
table today, look at the bottom of this email, a snapshot I capture from
my own IPv6 BGP4+ feeds. Not only I see /48s but also /64s and anything
else you can imagine. Call it swamp, soup or anything you like it's here
already.


> I am going to strongly disagree on this point at this time. We
> don't know that there will /ever/ be a strong migration of users
> to IPv6. IPv6 may yet flop completely, be replaced by IPv8

I'm sure Jim Fleming will be pleased :-D

> or something before it ever reaches full deployment, or even
> always live side by side with IPv4.

I don't think that a complete flop is realistic. Too many people have
invested too much money for IPv6 to sink overnight. The most likely
scenario is that is will live side-by-side with IPv6 for some years, and
eventually share the same fate as ISDN: I Still Don't Need and be
replaced by something else, the ISDN equivalent of cable or DSL (but not
IPv8 please).

The cost of enabling IPv6 in a Californian enterprise today is US$600
one-time and $100/yr recurring support cost until IPv4 is removed. This
is an un-scientific median cost from a small set of my own diverse
customers, but can't be _that_ much of course. Where's my ROI?

As far as end-customer demand, save for a massive "Intel Inside" like TV
campaign, I don't see how it could happen either. I will develop this
later as well, but IPv6 situation in terms of marketing is somehow
similar to microprocessor vendors: almost nobody buys a CPU alone these
days, and almost nobody will buy IPv6 as a standalone product either.
For the general public, "IPv6 inside" is the way to market. Catch: it
ain't no cheap.


> Even if we assume everything migrates to IPv6, I see no reason
> why we should change IPv4 policy at all.

Agree.

Michel.


"Interesting" announcements gathered a few minutes ago:
(alternatively, look at Jeroen Massar's stuff)

This might not look like a lot, but there are not quite 500 IPv6 routes
today, we're already looking at 12% of prefixes being announced being
soup, and this is _without_ any kind of IPv6 PI.

cisco7507#sh bgp ipv6 unicast
[removed as-paths and peers to protect the innocent]

*> 2001:288:3B0::/44
*> 2001:3C8:9009::/48
*> 2001:3C8:E109::/48
*> 2001:448:3::/48
*> 2001:460:410::/48
*> 2001:470:108::/48
*> 2001:470:110::/48
*> 2001:470:112::/48
*> 2001:470:1F00:954::/64
*> 2001:478::/45
*> 2001:478:65::/48
*> 2001:500::/48
*> 2001:500:1::/48
*> 2001:530:DEAD::/64
*> 2001:530:DEAD:BEAD::/64
*> 2001:570::/48
*> 2001:608:1::/48
*> 2001:608:6::/48
*> 2001:610:140::/48
*> 2001:610:240::/42
*> 2001:618:A::/48
*> 2001:628:14F9::/48
*> 2001:630:50::/48
*> 2001:648:2::/48
*> 2001:648:202::/48
*> 2001:660:3006::/48
*> 2001:700:FFFF::/48
*> 2001:730:2::/48
*> 2001:770:80::/48
*> 2001:7B8:200::/48
*> 2001:7F8:1::/48
*> 2001:7F8:2::/48
*> 2001:7F8:4::/48
*> 2001:7F8:5::/48
*> 2001:7F8:B::/48
*> 2001:7F8:18::/48
*> 2001:1458:E000::/48
*> 2001:1578:100::/40
*> 2001:1578:200::/40
*> 2001:1578:400::/40
*> 2002:C2B1:D06E::/48
*> 2002:C8A2::/33
*> 2002:C8C6:4000::/34
*> 2002:C8CA:7000::/36
*> 3FFE:4::/48
*> 3FFE:B00:4050::/48
*> 3FFE:1200:3028::/48
*> 3FFE:2022:F000::/48
*> 3FFE:2500:310::/48
*> 3FFE:2501:100::/48
*> 3FFE:2900:1:15::/64
*> 3FFE:2900:3::/48
*> 3FFE:2B00:1003::/48
*> 3FFE:3600:18::/48
*> 3FFE:4005:A::/48
*> 3FFE:400B:6002::/48
*> 3FFE:400B:6006::/48
*> 3FFE:8034:10::/48
*> 3FFE:80B0:1002::/48
*> 3FFE:8150:2001::/48
*> 3FFE:81D0:104::/48