rfc2072.txt

RFC 的详细文档！

          |
          |
          +---highOrderPart change only, high length change
          |
          |
          +---lowOrderPart change only, low length change
          |
          |
          +---totalPrefix change only, changes in high and low
          |
          |
          +---highOrderPart change only, no length change

   Ideally, a given prefix should either be "unchanged," "old," or
   "new." Renumbering will be easiest when each "old" prefix can be
   mapped to a single "new" prefix.



Berkowitz                    Informational                     [Page 11]

RFC 2072                Router Renumbering Guide            January 1997


   Unfortunately, the ideal often will not be attainable.  It may be
   necessary to run parts of the new and old address spaces in parallel.

   Renumbering applies first to prefixes and then to host numbers to the
   right of the prefix.  To understand the scope of renumbering, it is
   essential to:

      1.  Identify the prefixes (and possibly host fields) potentially
          affected by the renumbering operation.

      2.  Identify the authority that controls the values of the prefix,
          or part of the prefix, affected by renumbering.

   In a given enterprise, prefixes may be present that will be under the
   complete or partial control of the enterprise, as well as totally
   outside the control of the enterprise.  Let us review the principles
   of control over address space.

   More commonly, the most significant bits of the prefix are assigned
   to the enterprise by an address registry (e.g., InterNIC, RIPE, or
   APNIC) or by an Internet Service Provider (ISP).  This assignment of
   a value in the most significant bit positions historically has been
   called a "network number," when the assigned high-order part is 8,
   16, or 24 bits long.  More recent usage does not limit the assigned
   part to a byte boundary.  The preferred term for the assigned part is
   a "CIDR block" of a certain number of bits [RFC1518].

   The enterprise then extends the prefix to the right, creating
   "subnets."  It is critical to realize that routers make routing
   decisions based on the total prefix of interest, regardless of who
   controls which bits.  In other words, the router really doesn't know
   or care about subnet boundaries.

   The way to think about subnetting is that it creates a longer prefix.
   Even before CIDR, we collapsed multiple subnets into a single network
   number advertisement sent to external routers.  In a more general
   way, we now think of extending the prefix to the right as subnetting
   and collapsing it to the left as supernetting, aggregating, or
   summarizing.  Depending on the usage of subnetting or aggregation,
   different prefix lengths are significant at different router
   interfaces.

4.3  Renumbering Scope

   Prefixes may be taken from the private address space [RFC1918] that
   is not routable on the global Internet.  Since these addresses are
   not routable on the global Internet, changing parts of private
   address space prefixes is an enterprise-local decision.



Berkowitz                    Informational                     [Page 12]

RFC 2072                Router Renumbering Guide            January 1997


   If a prefix is totally outside the control of the enterprise, it is
   external, and will be minimally affected by routing.  Potential
   interactions of external prefixes with enterprise renumbering
   include:

      1)  Inadvertent alteration or deletion  of external addresses
          as part of router reconfiguration.
      2)  Loss of connectivity to application servers inside the
          enterprise, because the external client no longer knows
          the internal address of the server.
      3)  DNS/BGP
      4)  Security

   Prefixes partially under the control of the enterprise may change.
   The scope of this will vary depending on whether only the externally
   controlled part of the prefix changes, or if part of the internally
   controlled part is to be renumbered.  If the length of either the
   high-order or low-order parts change, the process becomes more
   complex.

   High-order-part-only renumbering is most common when an organization
   changes ISPs, and needs to renumber into the new provider's space.
   The old prefix may have been assigned to the enterprise but will no
   longer be used for global routing, or the old prefix may have been
   assigned to the previous provider.  Note that administrative
   procedures may be necessary to return the previous prefix, although
   this usually will be done by the previous provider.  There often will
   need to be a period of coexistence between the old and new prefixes.

   Low-order-part-only renumbering can occur when an enterprise modifies
   its internal routing structure, and the changes only affect the
   internal subnet structure of the enterprise network. This is typical
   of efforts involved in increasing the number of available subnets
   (e.g., for more point-to-point media) or increasing the number of
   hosts on a medium (e.g., in greater use of workgroup switches).

   Both the high-order and low-order parts may change.  This might
   happen when the enterprise changes to a new ISP, who assigns address
   space from a CIDR block rather than a classful network previously
   used.  With a different high-order prefix length, the enterprise
   might be forced to change its subnet structure.

5. Moving toward a Renumbering-Friendly Enterprise

   Renumbering affects both the configuration of specific router
   "boxes," and the overall system of routers in a routing domain.  The
   emphasis of this section is on making the current enterprise more
   renumbering-friendly, before any prefixes are actually changed.



Berkowitz                    Informational                     [Page 13]

RFC 2072                Router Renumbering Guide            January 1997


   Renumbering will have the least impact when the minimum number of
   reconfiguration options are needed.  When planning renumbering on
   routers, consider that many existing configurations may contain
   hard-coded IP addresses that may not be necessary, even if
   renumbering were not to occur.  Part of a router renumbering effort
   should include, wherever possible, replacing router mechanisms based
   on hard-coded addresses with more flexible mechanisms.

   Renumbering will also generally be easier if the configuration
   changes can be made offline on appropriate servers, and then
   downloaded to the router if the router implementation permits.

5.1  Default Routes

   A well-known method for reducing the amount of reference by one
   router to other routers is to use a default route to a higher-level,
   better-connected router.  This assumes a hierarchical network design,
   which is generally desirable in the interest of scaling.

   Default routes are most appropriate for stub routers inside a routing
   domain, and for boundary routers that connect the domain to a single
   ISP.

5.2  Route Summarization and CIDR

   When routes need to be advertised, summarize as much as is practical.
   Summarization is most effective when address prefixes have been
   assigned in a consistent and contiguous manner, which is often not
   the case in legacy networks.  Nevertheless, there is less to change
   when we can refer to blocks of prefixes.

   Not all routing mechanisms support general summarization.  Interior
   routing mechanisms that do include RIPv2, OSPF, EIGRP, IS-IS, and
   systems of static routes.  RIPv1 and IGRP do support classful
   summarization (i.e., at Class A/B/C network boundaries only).

   If existing addresses have been assigned hierarchically, it may be
   possible to renumber below the level of summarization, while hiding
   the summarization to the rest of the network.  In other words, if all
   the address bits being renumbered are to the right of the summarized
   prefix length, the change can be transparent to the overall routing
   system.

   Even when effective summarization is possible to hide the details of
   routing, DNS, filters, and other services may be affected by any
   renumbering.





Berkowitz                    Informational                     [Page 14]

RFC 2072                Router Renumbering Guide            January 1997


5.3  Server References in Routers

   Routers commonly communicate with an assortment of network management
   and other infrastructural servers.  Examples of these servers are
   given in the "Network Management" section below.  DNS itself,
   however, may be an important exception.

   Wherever possible, servers should be referenced by DNS name rather
   than by IP address.  If a specific router implementation only
   supports explicit address  references, this should be documented as
   part of the renumbering  plan.

   Routers may also need to  forward end host broadcasts to other
   infrastructure services (e.g., DNS, DHCP/BOOTP).  Configurations that
   do this are likely to contain hard-coded IP addresses of the
   destination hosts or their subnets, which will need to be changed as
   part of renumbering.

5.4  DNS and Router Renumbering

   The Domain Name Service is a powerful tool in any renumbering effort,
   and can help routers as well as end hosts.  If traceroute displays
   DNS names rather than IP addresses, certain debugging options can be
   transparent through the address transition.

   Be aware that dynamically learned names and addresses may be cached
   in router tables.  For a router to learn changes in address to name
   correspondence, it may be necessary to restart the router or
   explicitly clear the cache.

   Alternatively, router configuration files may contain hard-coded
   address/name correspondences that will not be affected by a change in
   the DNS server.

   Different DNS databases are affected by renumbering.  For example,
   the enterprise usually controls its own "forward" data base, but the
   reverse mapping data base may be maintained by its ISP.  This can
   require coordination when changing providers.

   Commonly, router renumbering goes through a transition period.
   During this transition, old and new addresses may coexist in the
   routing system.  Coexistence over a significant period of time is
   especially likely for DNS references to addresses that are known in
   the global Internet [deGroot].  Various DNS servers throughout the
   world may cache addresses for periods of days.






Berkowitz                    Informational                     [Page 15]

RFC 2072                Router Renumbering Guide            January 1997


   If, for example, a given router interface may have a coexisting new
   and old address, it can be appropriate to introduce the new address
   as an additional A record for the new address.

   DNS RR statements can end with a semicolon, indicating the rest of
   the line is a comment.  This can be used as the basis of tools to
   renumber DNS names for router addresses, by putting a comment (e.g.,
   ";newaddr") at the end of the A statements for the new addresses.  At
   an appropriate time, a script could generate a new zone file in which
   the new addresses become the primary definitions on A records, and
   the old addresses could become appropriately commented A records.  At
   a later time, these commented entries could be removed.

   Care should be taken to assure that PTR reverse mapping entries are
   defined for new addresses, because some router vendor tools depend on
   reverse mapping.

5.5  Dynamic Addressing

   Renumbering is easiest when addresses need to be changed in the least
   possible number of places.  Dynamic address assignment is especially
   attractive for end hosts, and routers may play a key role in this
   process.  Routers may act as servers and actually assign addresses,
   or may be responsible for forwarding end host address assignment
   requests to address assignment servers.

   The most common use of dynamic address assignment is to provide IP
   addresses to end systems.  Dynamic address assignment, however, is
   also used to assign IP addresses to router interfaces.  An address
   assignment server may assign an IP address to a router either in the
   usual DHCP way, based on a MAC address in the router, or simply based
   on the physical connectivity of the new router.  In other words, any
   router connected on a specific interface of the configuring router
   would be assigned the same IP address.

5.5.1 Router Roles in LAN-based DHCP Address Assignment

   End hosts attached to LANs often obtain address assignments from
   BOOTP or DHCP servers.  If the server is not on the same medium as
   the end hosts, routers may need to play a role in establishing
   connectivity between the end host and the address server.

   If the client is not on the same medium as the address assignment
   server, routers either must act as address assignment services, or
   forward limited broadcasts to the location of appropriate servers.