rfc2071.txt

著名的RFC文档,其中有一些文档是已经翻译成中文的的.

   Special cases of bridging are realized in workgroup switching
   systems, discussed below.

4.1.4  Limitations of Legacy Routing Systems

   Other performance problems might come from routing mechanisms that
   advertise excessive numbers of routing updates (e.g., RIP, IGRP).
   Likewise, appropriate replacement protocols (e.g., OSPF, EIGRP, S-IS)
   will work best with a structured addressing system that encourages
   aggregation.








Ferguson & Berkowitz         Informational                      [Page 5]

RFC 2071              Network Renumbering Overview          January 1997


4.1.5  Limitations of System Administration Methodologies

   There can be operational limits to growth based on the difficulty of
   adds, moves and changes.  As enterprise networks grow, it may be
   necessary to delegate portions of address assignment and maintenance.
   If address space has been assigned randomly or inefficiently, it may
   be difficult to delegate portions of the address space.

   It is not unusual for organizational networks to grow sporadically,
   obtaining an address prefix here and there, in a non-contiguous
   fashion.  Depending on the number of prefixes that an organization
   acquires over time, it may become increasingly unmanageable or demand
   higher levels of maintenance and administration when individual
   prefixes are acquired in this way.

   Reasonable IP address management may in general simplify continuing
   system administration; a good numbering plan is also a good
   renumbering plan.  Renumbering may force a discipline into system
   administration that will reduce long-term support costs.

   It has been observed "...there is no way to renumber a network
   without an inventory of the hosts (absent DHCP). On a large network
   that needs a database, plus tools and staff to maintain the
   database."[10] It can be argued that a detailed inventory of router
   configurations is even more essential.

4.2  Present

   Organizations now face needs to connect to the global Internet, or at
   a minimum to other organizations through bilateral private links.

   Certain new transmission technologies have tended to redefine the
   basic notion of an IP subnet.  An IP numbering plan needs to work
   with these new ideas. Legacy bridged networks and leading-edge
   workgroup switched networks may very well need changes in the
   subnetting structure.  Renumbering needs may also develop due to the
   characteristics of new WAN technologies, especially nonbroadcast
   multi-access (NBMA) services such as Frame-Relay and Asynchronous
   Transfer Mode (ATM).

   Increased use of telecommuting by mobile workers, and in small and
   home offices, need on-demand WAN connectivity, using modems or ISDN.
   Effective use of demand media often requires changes in numbering and
   routing.







Ferguson & Berkowitz         Informational                      [Page 6]

RFC 2071              Network Renumbering Overview          January 1997


4.2.1   Change in organizational structure or network topology

   As companies grow, through mergers, acquisitions and reorganizations,
   the need may arise for realignment and modification of the various
   organizational network architectures.  The connectivity of disparate
   corporate networks present unique challenges in the realm of
   renumbering, since one or more individual networks may have to be
   blended into a much larger architecture consisting a different IP
   address prefix altogether.

4.2.2  Inter-Enterprise Connectivity

   Even if they do not connect to the general Internet, enterprises may
   interconnect to other organizations which have independent numbering
   systems. Such connectivity can be as simple as bilateral dedicated
   circuits. If both enterprises use unregistered or private address
   space, they run the risk of using duplicate addresses.

   In such cases, one or both organizations may need to renumber into
   different parts of the private address space, or obtain unique
   registered addresses.

4.2.3   Change of Internet Service Provider

   As mentioned previously in Section 2, it is increasingly becoming
   current practice for organizations to have their IP addresses
   allocated by their upstream ISP.  Also, with the advent of Classless
   Inter Domain Routing (CIDR) [11], and the considerable growth in the
   size of the global Internet table, Internet Service Providers are
   becoming more and more reluctant to allow customers to continue using
   addresses which were allocated by the ISP, when the customer
   terminates service and moves to another ISP.  The prevailing reason
   is that the ISP was previously issued a CIDR block of contiguous
   address space, which can be announced to the remainder of the
   Internet community as a single prefix. (A prefix is what is referred
   to in classless terms as a contiguous block of IP addresses.)  If a
   non-customer advertises a specific component of the CIDR block, then
   this adds an additional routing entry to the global Internet routing
   table.  This is what is commonly referred to as "punching holes" in a
   CIDR block. Consequently, there are usually no routing anomalies in
   doing this since a specific prefix is always preferred over an
   aggregate route.  However, if this practice were to happen on a large
   scale, the growth of the global routing table would become much
   larger, and perhaps too large for current backbone routers to
   accommodate in an acceptable fashion with regards to performance of
   recalculating routing information and sheer size of the routing table
   itself.  For obvious reasons, this practice is highly discouraged by
   ISP's with CIDR blocks, and some ISP's are making this a contractual



Ferguson & Berkowitz         Informational                      [Page 7]

RFC 2071              Network Renumbering Overview          January 1997


   issue, so that customers understand that addresses allocated by the
   ISP are non-portable.

   It is noteworthy to mention that the likelihood of being forced to
   renumber in this situation is inversely proportional to the size of
   the customer's address space.  For example, an organization with a
   /16 allocation may be allowed to consider the address space
   "portable", while an organization with multiple non-contiguous /24
   allocations may not.  While the scenarios may be vastly different in
   scope, it becomes an issue to be decided at the discretion of the
   initial allocating entity, and the ISP's involved; the major deciding
   factor being whether or not the change will fragment an existing CIDR
   block and whether it will significantly contribute to the overall
   growth of the global Internet routing tables.

   It should also be noted that (contrary to opinions sometimes voiced)
   this form of renumbering is a technically necessary consequence of
   changing ISP's, rather than a commercial or political mandate.

4.2.3  Internet Global Routing

   Even large organizations, now connected to the Internet with
   "portable" address space, may find their address allocation too
   small. Current registry guidelines require that address space usage
   be justified by an engineering plan. Older networks may not have
   efficiently utilized existing address space, and may need to make
   their existing structures more efficient before new address
   allocations can be made.

4.2.4  Internal Use of LAN Switching

   Introducing workgroup switches may introduce subtle renumbering
   needs.  Fundamentally, workgroup switches are specialized, high-
   performance bridges, which make their main forwarding decisions based
   on Layer 2 (MAC) address information. Even so, they rarely are
   independent of Layer 3 (IP) address structure.  Pure Layer 2
   switching has a "flat" address space that will need to be renumbered
   into a hierarchical, subnetted space consistent with routing.

   Introducing single switches or stacks of switches may not have
   significant impact on addressing, as long as it is understood that
   each system of switches is a single broadcast domain. Each broadcast
   domain should map to a single IP subnetwork.

   Virtual LANs (VLANs) further extend the complexity of the role of
   workgroup switches. It is generally true that moving an end station
   from one switch port to another within the same VLAN will not cause
   major changes in addressing. Many overview presentations of this



Ferguson & Berkowitz         Informational                      [Page 8]

RFC 2071              Network Renumbering Overview          January 1997


   technology do not make it clear that moving the same end station
   between different VLANs will move the end station into another IP
   subnet, requiring a significant address change.

   Switches are commonly managed by SNMP applications. These network
   management applications communicate with managed devices using IP.
   Even if the switch does not do IP forwarding, it will itself need IP
   addresses if it is to be managed. Also, if the clients and servers in
   the workgroup are managed by SNMP, they will also require IP
   addresses. The workgroup, therefore, will need to appear as one or
   more IP subnetworks.

   Increasingly, internetworking products are not purely Layer 2 or
   Layer 3 devices. A workgroup switch product often includes a routing
   function, so the numbering plan must support both flat Layer 2 and
   hierarchical Layer 3 addressing.

4.2.4  Internal Use of NBMA Cloud Services

   "Cloud" services such as frame relay often are more economical than
   traditional services. At first glance, when converting existing
   enterprise networks to NBMA, it might appear that the existing subnet
   structure should be preserved, but this is often not the case.

   Many organizations often  began by treating the "cloud" as a single
   subnet, but experience has shown it is often better to treat the
   individual virtual circuits as separate subnets, which appear as
   traditional point-to-point circuits.  When the individual point-to-
   point VCs become separate subnets, efficient address utilization
   requires the use of long prefixes (i.e., 30 bit) for these subnets.
   In practice, obtaining 30 bit prefixes means the logical network
   should support variable length subnet masks (VLSM).  VLSMs are the
   primary method in which an assigned prefix can be subnetted
   efficiently for different media types. This is accomplished by
   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, and may be less
   reliable because a single routing point of failure is created.
   Motivations for such alternatives tend to include:








Ferguson & Berkowitz         Informational                      [Page 9]

RFC 2071              Network Renumbering Overview          January 1997


      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 "hub-and-spoke" implementations, static routing can
          be preferable from a performance and flexibility standpoint,
          since it does not produce routing protocol chatter and is
          unaffected by split horizon constraints (namely, the inability
          to build an adjacency with a peer within the same IP
          subnetwork).