BOF topic - address allocation

[Pedro Roque]

This is a proposal for a small modification of RFC 1897 which could be
considered a complement to what Alain Durand was proposing in a previous
mail. As 1897 is a formal document i thought i better write the "mail"
in a similiar way.

./Pedro.

------------------------------------------------------------------------------
Submission to the 6bone BOF					     Roque, P.
November 1996						     roque@di.fc.ul.pt

		   Address Allocation on the 6bone


  Abstract

    This document suggests an address allocation policy for use in the
    6bone.

  Introduction

    Currently, the 6bone, although composed of a small number of nodes
    starts to show some characteristics that should be avoided in terms
    of address allocation.

    The current 6bone topology, appears as a mixture of a full mesh network
    and a geographic based network. This fact tends to increase the
    complexity of the management tasks and routing table determination.
    But as routing protocols start to be deployed, the topology moves
    to a more common model of a large group of leaf domains and a smaller
    group of transit domains, the current address allocation [RFC1897] 
    will still force transit routers to carry an average of one entry
    per connected site.

    The goal of this document is to propose a way to reduce this ratio
    via changes to the address allocation procedure. Complementary, it
    is believed that it will reduce the management tasks of the network
    (specially until the point that every transit site speaks a common
    routing protocol).

  "Real" IPv6 addresses

    Using IPv6 addresses out of the "Test Address Allocation" space in
    the current 6bone is an alternative that the author considers to be 
    a potential disservice to IPv6 deployment.

    Of the current 6bone sites that provide for transit traffic a very
    small percentage are transit domains for the IPv4 Internet. Since
    an address allocation should distribute small prefixes for transit
    domains for further allocation to leaf nodes that default through
    them, it would be necessary to allocate at the moment "top level
    provider" prefixes to the transit domains.

    Since the significative majority of those domains will not, under
    normal circumstances, be considered "top level" transit networks
    this policy will then to waste address space and reduce the aggregation
    capability of the future IPv6 Internet.

    The author believes that an address allocation policy should follow
    the wire topology, and not a virtual topology like the current 6bone.

  Modifications to the Test Address Allocation

    RFC 1897 specifies that a site should use a prefix composed of the
    test allocation prefix (5f::/8), it's network provider AS number
    and it's IPv4 network addresses.

    The author suggests that for leaf nodes, the prefix should be
    constructed using the 32 first bits of the prefix used by the
    transit site they connect to. The following 32 bits of the prefix
    should be constructed according to the rule set on the referred
    document.

    For transit domains, the AS number used should be an AS number
    registered in the InterNIC database for that domain, if available.
    Else the transit domain will use the AS number of it's network
    provider and a sequential number (starting from 1) on the first
    reserved field of RFC 1897 (from left to right).

    It is expected that users of the same AS can resolve the attribution
    of the sub-AS identifier as a local problem.

    As the 6bone grows it's topology is expected to converge to the current
    Internet topology and transit sites will normally own an AS identifier.

  Lifetime

    This policy is only intended to the short term, and if adopted, should
    be reviewed periodicly.