INTERNET DRAFT M. Ohta
draft-ohta-ipv6-addr-arch-00.txt Tokyo Institute of Technology
March 1995
Provider Independent IPv6 Addressing Architecture
Status of this Memo
This document is an Internet-Draft. Internet-Drafts are working
documents of the Internet Engineering Task Force (IETF), its areas,
and its working groups. Note that other groups may also distribute
working documents as Internet-Drafts.
Internet-Drafts are draft documents valid for a maximum of six
months. Internet-Drafts may be updated, replaced, or obsoleted by
other documents at any time. It is not appropriate to use Internet-
Drafts as reference material or to cite them other than as a
``working draft'' or ``work in progress.''
To learn the current status of any Internet-Draft, please check the
1id-abstracts.txt listing contained in the Internet-Drafts Shadow
Directories on ds.internic.net, nic.nordu.net, ftp.nisc.sri.com, or
munnari.oz.au.
Abstract
An IPv6 addressing architecture which maximize the provider
independence is described.
With flatly routed IPv4 addresses, one can subscribe to multiple
providers and change providers at will without a lot of efforts.
But, IPv6 packets will be hierarchically routed and their addresses
will have hierarchical structure, whose higher part is determined by
the network provider.
By separating a 128 bit IPv6 address into 64 bit ILOC (Internet
LOCator), first 4 bytes of which is flat routable provider part and
the rest 4 bytes of which is hierarchical intra-provider part, and 64
bit IID (Internet ID), which is not routable but globally unique, it
is possible to preserve some of the provider independence of IPv4.
The architecture can also identify geographical location of
providers.
Introduction
IPv4 addresses do not contain provider dependent information. Thus,
with IPv4 addresses, we can select and change providers without
M. Ohta Expires on Sept. 25, 1995 [Page 1]
INTERNET DRAFTProvider Independent IPv6 Addressing Architecture March 1995
reassigning IP addresses.
On the other hand, IPv6 addresses contain provider dependent part for
routing aggregation.
If host identification is controlled by providers, it is difficult to
change providers or have multiple provider. It is likely that
subscribers are tuned into the provider they choose and tends to
promote provider monopoly.
The problem, obviously, is in provider-dependent hierarchical
routing.
And, the solution, obviously, is not to do provider-dependent
hierarchical routing. This memo proposes an address assignment
scheme for IPv6 which does flat routing for providers.
Routing below providers is, of course, hierarchical so that enough
scalability is assured to support 10^12 networks and 10^15 hosts.
A mechanism to identify small geographical locations is also included
to have geographically-near-optimal and least-costly routing with
proper route selection.
Assignment Plan for IPv6 address
The 16 byte IPv6 address is divided into two fields: 8 byte ILOC
(Internet LOCator) and 8 byte IID (Internet ID).
ILOC is further divided into three sub fields: 4 byte "Provider ID",
2 byte "Subscriber ID" and 2 byte "Subnet ID".
"Provider ID" and "Subscriber ID" together is called "Provider
dependent part".
Provider dependent part is supplied by providers and dynamically
reconfigurable at system boot time or even during operation. It is
expected that routers to providers announce provider dependent part
information.
IID is a globally unique ID of a host or a multicast group to
identify the host or the multicast group. While an IID uniquely
identifies a single host, a host may have multiple IIDs. But, within
a lifetime of some connection or reservation such as for TCP or flow,
the same IID should be used regardless of the routing changes.
IID is supplied by subscribers. The configuration may be automatic.
But it is expected that renumbering is necessary not so often, in
M. Ohta Expires on Sept. 25, 1995 [Page 2]
INTERNET DRAFTProvider Independent IPv6 Addressing Architecture March 1995
general, only when a host is purchased or the host is moved to
different suborganization of the provider. Host specific information
such as IP address to host name mapping is looked up only through
IID. For autoconfiguration, it must be possible to derive some IIDs
from MAC addresses. As 2^28<10^15, MAC addresses to support 10^15
hosts must be longer than 48 bits (actually, a lot longer than that),
IID is designed to have 64 bits.
Subnet ID is also supplied by subscribers and identifies a subnet
within a subscribers LAN. The configuration may be automatic through
nearest routers. Renumbering is necessary when a location of a host
is changed to a different subnet.
Network layer identification of a host is done through IID just like
the current IPv4.
Routing is controlled purely by the ILOC part of IPv6 address, which
is 8byte aligned and, thus, is more efficient than schemes using full
16 bytes.
For the deliverly between adjacent routers or a router and a host,
which is within a single link layer, only the identification is
necessary. So, for IPv6 ARP equivalent, IID is used and ILOC is not
consulted.
Users can change providers at will just by disconnecting one of its
external routers and connect it to a new provider, and ILOC part will
be automatically reconfigured. But, it is still necessary to update
information of DNS, which is further explained in the "DNS
interaction" section.
A host of a subscriber belonging to multiple providers may have
multiple provider dependent parts.
Different interfaces of a host is, in general, distinguished by ILOC
part.
Assignment Plan for Provider ID, Subscriber ID and Subnet ID
4 byte provider ID uniquely identifies a single provider in the
Internet, while a provider may have multiple provider IDs.
4 byte provider ID combined with 2 byte subscriber ID uniquely
identifies a subscriber in the Internet.
Provider ID of 0 is reserved for subscriber local routing.
2 byte subnet ID uniquely identifies a subnet in a subscribers
M. Ohta Expires on Sept. 25, 1995 [Page 3]
INTERNET DRAFTProvider Independent IPv6 Addressing Architecture March 1995
network.
A large subscriber having more than 65536 subnets will have multiple
subscriber IDs from a provider.
A large provider having more than 65536 subscriber IDs or having some
geographical constraints, as explained in the next section, will have
multiple provider IDs.
Avoiding Treework
A goal of IPv6 is to avoid treework and make it real network.
With usual provider based addressing, users within a single provider
share the same routing prefix so that it is difficult to use better
route outside the provider. That is, complex configuration is
necessary to break the routing hierarchy, which does not scale. It
is unlikely that providers welcome such configuration only to allow
subscribers pay less to the provider.
So, in this proposal, to identify small geographical locations, a
provider ID should not cover an area of 100Km radius. That is, a
large provider must use different provider IDs for hosts located more
than 200Km apart. The distance is measured only for IP entry point
of providers, so that PPP service through non-IP public data network
for 1,000Km-distant user is allowed.
Thus, it becomes naturally possible to favor local links outside of
the centralized provider, if local IXes are available.
Only about 17,000 IDs are necessary to cover the surface of the
Earth. Inter-planetary communication is NOT considered here.
Assignment Plan for IIDs
There are two kinds of IIDs, structured and unstructured.
Structured IIDs are maintained by IANA and is used to lookup IANA
maintained database of in-addr.arpa. equivalent of IPv6.
Unstructured IIDs are directly derived from globally unique MAC
addresses of hosts and useful for autoconfiguration.
The most significant bit of structured IID is 0.
First three bytes of structured IID are assigned from IANA to country
NICs.
M. Ohta Expires on Sept. 25, 1995 [Page 4]
INTERNET DRAFTProvider Independent IPv6 Addressing Architecture March 1995
Each country NIC uses the next three bytes for independent
subscribers.
A subscriber use the last two bytes for the internal use.
If the lease significant bit of IID is 0, it is for unicasting.
Otherwise, the IID identifies multicast. An IID of all 1's except of
the first bit is for broadcasting.
The most significant bit of unstructured IID is 1. For the current 48
bit MAC addresses maintained by IEEE, 16 bits prefix of
"1000000000000000" is added to form the IID. Rest of the unstructured
IID space is reserved for the future use.
DNS Interaction
While IPv6 addresses are mostly autoconfigurable, it is still
necessary to use DNS to advertise IPv6 addresses all over the
Internet.
As IIDs are considered to be rather static, they can be stored in DNS
just as the current IPv4 addresses.
On the other hand, if a subscriber adds or changes a provider, it is
necessary to reflect the change to DNS. To do so without a lot of
pain, the DNS lookup of ILOC should be indirect. That is, each DNS
node of a host should not directly have ILOC but have a pointer to
some node. The node, which is pointed by a lot of hosts within the
same subscriber, then, have (multiple) ILOC information of the
subscriber. Thus, it is possible to quickly change a provider without
much difficulty. Of course, some statically configured raw addresses,
which is necessary for the minimal operation of DNS itself, will
still need to be reconfigured.
Supporting 10^12 networks and 10^15 hosts
How the requirement to support 10^12 networks and 10^15 hosts can be
satisfied?
First, how routing between 10^12 networks is possible? 10^3 hosts
within a subnet can easily be identified by the IID. Thus, (Provider
ID, Subscriber ID, Subnet ID) must identify 10^12 networks. It is
not unnatural that a provider, in average, supports at least 10^3
subscribers. It can also be safely assumed that subnet ID can
identify 10^1 subnets. Thus, Provider ID is required to identify
10^8 hosts. Considering that the requirement contains 10^2 safety
factor, the least significant byte of the Provider ID are reserved
for the factor. The remaining 3 bytes (2^24>10^7) are much more than
M. Ohta Expires on Sept. 25, 1995 [Page 5]
INTERNET DRAFTProvider Independent IPv6 Addressing Architecture March 1995
enough to identify 10^6 providers. It is assumed that by the time we
support 10^12 networks, flat routing of 10^6 providers is not a
problem at all. The reserved lowest byte of provider ID may also be
used for two-level routing among providers.
Next, how identification of 10^15 hosts is possible? 10^3 hosts of a
subscriber can easily be identified by the last two bytes of IID.
For the remaining 10^12 factor, IANA and country NICs are expected to
manage the upper 6 bytes (for about 2*10^24 hosts) densely enough.
Thus, it is feasible that more than 10^12 networks can be identified
with the 6 bytes.
Conclusion
As the 16 byte address space is so large, it is possible to use it
wisely to enjoy full provider independence, including provider change
without renumbering and long distance provider selection by provider
IDs.
Acknowledgement
The work is inspired by the concept of locator/EID of NIMROD and PIP.
Josh Osborne helped the Author to clarify the geographical
identification feature of the proposal.
References
(to be provided)
Security Considerations
(to be provided)
Author's Addresses
Masataka Ohta
Computer Center
Tokyo Institute of Technology
2-12-1, O-okayama, Meguro-ku
Tokyo 152, JAPAN
Phone: +81-3-5734-3299
Fax: +81-3-5734-3415
EMail: mohta@necom830.cc.titech.ac.jp
M. Ohta Expires on Sept. 25, 1995 [Page 6]