Re: [GLOBAL-V6] PI space under IPv6

[Jeroen Massar <jeroen@unfix.org>]

Brian E Carpenter wrote:
> 
> On 2007-01-17 02:20, Jeroen Massar wrote:
[..]
>> No, the IPv6 routing table will include the IPv6 routing table.

Hmm, I guess I meant to write there "IPv6 ... will include IPv4 ..." ;)

> Actually that was one of the main concerns when we designed 6to4 -
> that if badly deployed, it would import the whole IPv4 swamp
> (a.k.a. PI space) into the IPv6 routing table. That's why RFC 3056
> explicitly forbids it:

Indeed, but it seems that those rules are correctly adhered to
fortunately: http://www.sixxs.net/tools/grh/lg/?when=now&find=2002::/16

The big problem with 6to4 is anyway that there are not too many anycast
deployments, especially after that German ISP sending several hundreds
of mbits to them (they couldn't care they claimed and the receiving ends
should upgrade if they couldn't handle a bit of traffic), causing
several of the ISP's hosting anycast relays to shut them down as it is
for them a free service to the community.

Next to that the other problem is proto-41 doesn't easily traverse NAT's
which is especially the case with 6to4 due to the address being inside
the protocol. One can hack that up a bit to make it work but that is far
from viable for most users.

Btw, did CERNET2 drop off the Internet? It is in BGP, but apparently
Hurricane Electric is their main 'transit' and that is where also a lot
of packets go missing. I can reach CERNET from SWITCH in Switzerland,
going over GEANT, but any path over HE.net seems to be broken ;(
(see below for some traces)


JORDI PALET MARTINEZ wrote:
> We need better ways to do routing, and several groups, including IETF
> are already working on that. Meanwhile, I think all the regions should
> support IPv6 PI so they aren’t in disadvantage versus ARIN and APNIC
> service regions (which are the ones now that have IPv6 PI policies).

As Roger also mentioned in his mail, there should not be a distinction
between "PI" and "PA" at all. Like I mentioned in my mail, RIR's should
care about providing the required addresses to the organizations that
need them, they should not care if that takes up one or 1000 routing slots.

What RIR's do have to look at it that they give organizations a
reasonable size of address space and not wasting it. Giving a single
person a /32 for instance is very wasteful, giving that to an
organization that can justify it, isn't wasteful. There is enough
address space, but still, that was also the case when the A-classes
where being divided; fortunately with IPv6 there are always 8 tries ;)

Greets,
 Jeroen

--

Most routes I tried, which go over HE.net break (note in BGP the next
hop is HE.net / AS6939, which does not respond or something):

traceroute to www.cernet2.net (2001:250:c006::3) from
2001:838:1:1:210:dcff:fe20:7c7c, 30 hops max, 16 byte packets
 1  fe0.breda.ipv6.concepts-ict.net (2001:838:1:1::1)  1.068 ms  0.506
ms  0.461 ms
 2  se2.ams-ix.ipv6.concepts-ict.net (2001:838:0:10::1)  2.365 ms  3.975
ms  3.243 ms
 3  nl-ams04a-re1-fe-0-0.ipv6.aorta.net (2001:7f8:1::a500:6830:1)  6.635
ms  6.556 ms  5.3 ms
 4  nl-ams06d-re1-t-2.ipv6.aorta.net (2001:730::1:c)  5.5 ms  9.603 ms
9.534 ms
 5  hurrican.net-gw1.nl.ipv6.aorta.net (2001:730::1:2f)  170.477 ms
170.3 ms  169.274 ms
 6  * * *
 7  * *


From SWITCH works:

 4  swiCE3-10GE-1-1.switch.ch (2001:620:0:c027::1)  4.233 ms  4.162 ms
4.329 ms
 5  swiCE2-10GE-1-4.switch.ch (2001:620:0:c03d::1)  4.486 ms  4.396 ms
4.426 ms
 6  switch.rt1.gen.ch.geant2.net (2001:798:12:10aa::1)  4.479 ms  4.477
ms  4.455 ms
 7  so-7-2-0.rt1.fra.de.geant2.net (2001:798:cc:1201:1401::2)  12.622 ms
 12.597 ms  12.591 ms
 8  so-1-3-0.rt1.lux.lu.geant2.net (2001:798:cc:1401:1501::2)  26.669 ms
 26.68 ms  26.678 ms
 9  bj-so-02-622m.bb-v6.noc.tein2.net (2001:254:1:6::1)  167.408 ms
167.006 ms  167.084 ms
10  cn.pr-v6.noc.tein2.net (2001:254:1:7::2)  166.99 ms  167.053 ms
166.945 ms
11  r0v-r-bj.v6.cernet.net (2001:250:0:3::2)  307.664 ms  239.222 ms
256.222 ms
12  2001:250:c000:20::1 (2001:250:c000:20::1)  167.494 ms  167.457 ms
167.65 ms
13  www.6tb.net (2001:250:c006::3)  167.854 ms  167.914 ms  167.721 ms

That last '6tb.net' domain is odd as it was acquired by a domainsquatter
in oct 2006.

Attachment: signature.asc
Description: OpenPGP digital signature