[sig-policy] Fwd: IPv6 address assignment proposal
All,
This is another proposal about IPv6 assignment to the Root
DNS servers from Akira Kato and Jun Murai of WIDE project.
This is a little bit overlapped by Okutani-san's proposal.
It might be better to discuss about these two proposals
in this mailing list before the meeting.
Regards,
Takashi Arano
> Akira Kato and Jun Murai
> WIDE Project
> draft on Jul 28, 2002
>
> Proposal for IPv6 Address Assignment to Root DNS Servers
>
>
>1. Introduction
>
>Current practice for assignment of the IPv6 addresses is based on
>a notion "aggregatable" in which (relatively) large ISPs obtain the
>address blocks from one of the Regional Internet Registries and delegate
>the subspace to their customer ISPs or to their direct end customers in
>order to accomodate maximum routing information aggregation. So the
>current IPv6 addresses are not considered portable -- if a site (an
>enterprise or an ISP) changes its upstream ISP, the IPv6 address will
>be changed.
>
>In order to provide IPv6 transport support for the Root DNS servers,
>they need IPv6 addresses. These IPv6 addresses will also be described
>in the "hist" files which will eventually be distributed to virtually
>all of the DNS servers. Therefore, renumbering the IPv6 addresses of
>the Root DNS servers is practically impossible. So the "portable" IPv6
>addresses are necessary for the Root DNS servers.
>
>
>2. Proposal
>
>While each Root DNS Server requires only a few IPv6 addresses (in an
>extreme case only single IPv6 service address is required), each Root DNS
>Server should be assigned a /32 (or standard size at the time of assignment)
>from corresponding RIR's.
>
>RIPE/NCC has similar program descried in RIPE-233 (see http://www.ripe.net/
>ripe/docs/ipv6-rootservers.html).
>
>The assigned IPv6 addresses should be used only for the Root DNS service
>and must not be used with other services unless otherwise approved by
>RSSAC.
>
>The Root DNS Servers may use Anycast as described in RFC3258. While the
>usage of anycast model on the real Root DNS Servers is up to the Root
>DNS Server operators' community or to ICANN RSSAC, it may possible to
>use the IPv6 prefix assigned by one RIR to the server whose geographical
>region is out side of the corresponding RIR's service area. This may be
>preferred to maximize the resiliency as well as to minimize the round
>trip time. The RIR should agree with this type of usage even when all of
>the Root DNS Servers using the particular IPv6 address move outside of
>the RIR's service area. When the particular IPv6 address no longer used,
>the address must be returned to the original RIR by the Root DNS Server
>operators' community.
>
>
>3. Consideration
>
>It may possible that a common prefix for IPv6 PI address block is
>defined and that each Root DNS server is assigned a small chunk (such as
>/48 or /64) from the block. The micro allocation policy is one of the
>variants of the IPv6 PI address. The IPv6 address assigned by the micro
>allocation policy is not considered routable. While it may be suitable
>for IPv6 Internet exchange points, it is not applicable to the Root DNS
>servers where routability is extremely important.
>
>It may be possible to encourage the IPv6 ISPs to relax the IPv6 prefix
>filtering rule defined by the current routing practice so that they make
>those micro allocated IPv6 addresses to the Root DNS servers routable
>while others are filtered out.
>
>But if some of the transit ISPs fail to implement the special filtering
>rule, some Root DNS servers may not be reachable beyond those ISPs.
>If all of the Root DNS servers are depending upon the special filtering
>rule, some portion of the IPv6 Internet will loose the access to all of
>the Root DNS Servers and their customers will not able to use DNS at all.
>
>Therefore it is suggested that all (or most of) Root DNS Server should
>use prefixes with the regular length scattered across the regular allocation
>space rather than to use special prefix length and special address space.
>
>The proposed scheme will contribute the growth of the routing table.
>There are 13 Root DNS servers, currently. Due to the packet size
>limitation, it is possible to add only up to two AAAA records to the
>"root-servers.net" zone. IPv6 unfragmented UDP packet size can grow up
>to 1280bytes. When EDNS0 is widely deployed in those DNS servers,
>a DNS packet may include up to 20 NS records for root each with an A
>record and an AAAA record. When the transition to IPv6 is completed
>and no IPv4 support is necessary, it is possible to include up to 28
>Root DNS Servers. If all of the Root DNS Servers will use the proposed
>address space, the maxmum contribution to the global IPv6 routing table
>will be 28 and it will be only a small portion of the entire routing table.
>
>Authors' addresses
>
> Akira Kato
> WIDE Project
> The University of Tokyo, Information Technology Center
> 2-11-16 Yayoi Bunkyo Tokyo
> 113-8658, JAPAN
> Tel: +81 3-5841-2726
> Email: kato at wide dot ad dot jp
>
> Jun Murai
> Keio University, Faculty of Environmental Informations
> 5322 Endo Fujisawa Kanagawa
> 252-8520 JAPAN
> Tel: +81 466-49-1100
> Email: junsec at wide dot ad dot jp
* sig-policy: APNIC SIG on resource management policy *
* To unsubscribe: send "unsubscribe" to sig-policy-request at apnic dot net *