rfc2072.txt

RFC 的详细文档！

   prefix lengths, establishing one or more prefix lengths for LAN media
   with more than two hosts, and subdividing one or more of these
   shorter prefixes into longer /30 prefixes that minimize address loss.

   There are alternative ways to configure routing over NBMA, using
   special mechanisms to exploit or simulate point-to-multipoint VCs.
   These often have a significant performance impact on the router, and
   may be less reliable because a single point of failure is created.
   Mechanics of these alternatives are discussed later in this section,
   but the motivations for such alternatives tend to include:

      1.  A desire not to use VLSM.  This is often founded in fear
          rather than technology.
      2.  Router implementation issues that limit the number of subnets
          or interfaces a given router can support.
      3.  An inherently point-to-multipoint application (e.g., remote
          hosts to a data center).  In such cases, some of the
          limitations are due to the dynamic routing protocol in use.
          In such "star" applications, static routing may actually be
          preferable from performance and flexibility standpoints,
          since it does not produce routing traffic and is unaffected
          by split horizon.

   To understand how use of NBMA services affects the addressing
   structure and routers, it is worth reviewing what would appear to be
   very basic concepts of IP subnets.  The traditional view is that a
   single subnet is associated with a single physical medium.  All hosts
   physically connected to this medium are assumed to be able to reach
   all other hosts on the same medium, using data link level services.
   These services are medium specific:  hosts connected to a LAN medium
   can broadcast to one another, while hosts connected to a point-to-
   point line simply need to transmit to the other end.






Berkowitz                    Informational                      [Page 6]

RFC 2072                Router Renumbering Guide            January 1997


   When one host desires to transmit to another, it first determines if
   the destination is local or remote.  A local destination is on the
   same subnet and assumed to be reachable through data link services.
   A remote destination is on a different subnet, and it is assumed that
   router intervention is needed to reach it.

   The first NBMA problem comes up when a single subnet is implemented
   over an NBMA service.  Frame Relay provides single virtual circuits
   between hosts that have connectivity.  It is quite common to design
   Frame Relay services as partial meshes, where not all hosts have VCs
   to all others.  When the set of hosts in a partial mesh is in a
   single IP subnet, partial mesh violates the local model of full
   connectivity.  Even when there is full meshing, a pessimistic but
   reasonable operational model must consider that individual VCs do
   fail, and full connectivity may be lost transiently.

   There are several ways to deal with this violation, each with their
   own limitations.  If a specific "central" host has connectivity to N
   all other hosts, that central host can replicate all frames it
   receives from one host onto outgoing VCs connecting it with the (N-1)
   other hosts in the subnet.  Such replication usually causes an
   appreciable CPU load in the replicating router.   The replicating
   router also is a single point of failure for the subnet.  This method
   does not scale well when extended to fuller meshes within the subnet.

   In a routing protocol, such as OSPF, that has a concept of designated
   routers, explicit configuration usually is needed.  Other problems in
   using a meshed subnet is that all VCs may not have the same
   performance, but the router cannot prefer individual paths within the
   subnet.

   One of the simplest methods is not to attempt to emulate a broadcast
   medium, but simply to treat each VC as a separate subnet.  This will
   cause a need for renumbering.  Efficient use of the address space
   dictates a /30 prefix be used for the per-VC subnets.  Such a prefix
   often needs VLSM support in the routers.

3.4  Expansion of Dialup Services

   Dialup services, especially public Internet access providers, are
   undergoing explosive growth.   This success represents a particular
   drain on the available address space, especially with a commonly used
   practice of assigning unique addresses to each customer.








Berkowitz                    Informational                      [Page 7]

RFC 2072                Router Renumbering Guide            January 1997


   In this practice, individual users announce their address to the
   access server using PPP's IP configuration option [RFC1332].  The
   server may validate the proposed address against some user
   identifier, or simply make the address active in a subnet to which
   the access server (or set of bridged access servers) belongs.

   These access server functions may be part of the software of a
   "router" and thus are within the scope of this Guide.

   The preferred technique [Hubbard] is to allocate dynamic addresses to
   the user from a pool of addresses available to the access server.
   Various mechanisms are used actually to do this assignment, and are
   discussed in Section 5.5.

3.5  Internal Use of Switched Virtual Circuit Services

   Services such as ATM virtual circuits, switched frame relay, etc.,
   present challenges not considered in the original IP design.  The
   basic IP decision in forwarding a packet is whether the destination
   is local or remote, in relation to the source host's subnet.  Address
   resolution mechanisms are used to find the medium address of the
   destination in the case of local destinations, or to find the medium
   address of the router in the case of remote routers.

   In these new services, there are cases where it is far more effective
   to "cut-through" a new virtual circuit to the destination.  If the
   destination is on a different subnet than the source, the cut-through
   typically is to the egress router that serves the destination subnet.

   The advantage of cut-through in such a case is that it avoids the
   latency of multiple router hops, and reduces load on "backbone"
   routers.  The cut-through decision is usually made by an entry router
   that is aware of both the routed and switched environments.

   This entry router communicates with a address resolution server using
   the Next Hop Resolution Protocol (NHRP) [Cansever] [Katz].  This
   server maps the destination network address to either a next-hop
   router (where cut-through is not appropriate) or to an egress router
   reached over the switched service.  Obviously, the data base in such
   a server may be affected by renumbering.  Clients may have a hard-
   coded address of the server, which again may need to change.

   While the NHRP work is in progress at the time of this writing,
   commercial implementations based on drafts of the protocol standard
   are in use.






Berkowitz                    Informational                      [Page 8]

RFC 2072                Router Renumbering Guide            January 1997


4.  Numbering and Renumbering

   What is the role of any numbering plan?  To understand the general
   problem, it can be worthwhile to review the basic principles of
   routers.  While most readers will have a good intuitive sense of
   this, the principles have refined in the current usage of IP.

   A router receives an inbound IP datagram on one of its interfaces,
   and examines some number of bits of the destination address.  The
   sequence of bits examined by the router always begin at the left of
   the address (i.e., the most significant bit).  We call this sequence
   a "prefix."

   Routing decisions are made on totalPrefix bits, which start at the
   leftmost (i.e., most significant) bit position of the IP address.
   Those totalPrefix bits may be completely under the control of the
   enterprise (e.g., if they are in the private address space), or the
   enterprise may control the lowOrderPrefix bits while the
   highOrderPrefix bits are assigned by an outside organization.

   The router looks up the prefix in its routing table (formally called
   a Forwarding Information Base).  If the prefix is in the routing
   table, the router then selects an outgoing interface that will take
   the routed packet to the next hop IP address in the end-to-end route.
   If the prefix cannot be found in the routing table, the router
   returns an ICMP Destination Unreachable message to the source address
   in the received datagram.

   Assuming the prefix is found in the routing table, the router then
   transmits the datagram through the indicated outgoing interface. If
   multicast routing is in effect, the datagram may be copied and sent
   out multiple outgoing interfaces.



















Berkowitz                    Informational                      [Page 9]

RFC 2072                Router Renumbering Guide            January 1997


4.1  Categorizing the topology

   From the router renumbering perspective, renumbering impact is apt to
   be greatest in highly connected parts of "backbones," and least in
   "stub" parts of the routing domain that have a single route to the
   backbone.

                         Global Internet
                            ^
                            |
                            |
                          Back1-------------------Back2
                            |                       |
                      +-----------+              +----------+
                      |           |              |          |
                    Reg1.1------Reg1.2          Reg2.1-----Reg2.2
                    |           |               |          |
                    |           |               |          |
                  Branch       Branch         Branch      Branch
                  1.1.1 to     1.2.1 to       2.1.1 to    2.2.1 to
                  1.1.N        1.2.N          2.1.N       2.2.N

   In this drawing, assume Back1 and Back2 exchange full routes; Back1
   is also the exterior router.  Regional routers (Reg) exchange full
   routes with one another and aggregate addresses to the backbone
   routers.  Branch routers default to regional routers.

   From a pure topological standpoint, the higher in the hierarchy, the
   greater are apt to be the effects of renumbering.  This is a first
   approximation to scoping the task, assuming addresses have been
   assigned systematically.  Systematic address space is rarely the case
   in legacy networks.



















Berkowitz                    Informational                     [Page 10]

RFC 2072                Router Renumbering Guide            January 1997


4.2  Categorizing the address space

   An inventory of present and planned address space is a prerequisite
   to successful renumbering.  Begin by identifying the prefixes in or
   planned into your network, and whether they have been assigned in a
   systematic and hierarchical manner.

       +--Unaffected by renumbering [A]
       |
       |
       +--Existing prefixes to be renumbered
       |  |
       |  |
       |  +----To be directly renumbered on "flag day"
       |  |
       |  |
       |  +----Initially to be renumbered to temporary address
       |
       |
       +--Existing prefixes to be retired
       |
       |
       +--Planned new prefixes
          |
          |
          +---totalPrefix change, no length change
          |
          |
          +---highOrderPart change only, no length change
          |
          |
          +---lowOrderPart change only, no length change