rfc2791.txt

著名的RFC文档,其中有一些文档是已经翻译成中文的的.

      This method includes the transit core and each regional network
      into one AS domain. The routing hierarchy is realized by utilizing
      multi-level IS-IS or OSPF areas and either BGP Confederation or
      I-BGP Reflector or a combination of the two.

      This mechanism avoids the introduction of an extra AS in the
      routing path, which is an advantage over the method described in
      Point 1).  However, multi-area hierarchical IGP is rarely used
      now-a-days in large networks since most of them are using IS-IS
      for internal routing, which does not have sufficient multi-level
      support. Although IS-IS supports multi-area routing, it imposes a
      strict hierarchy between backbone and sub-areas and allows only
      the advertisement of a default route from the backbone area to the
      sub-areas instead of specific prefixes. This restriction may be
      suitable for a network with a simple sub-area topology. A sub-area
      in a large network, typically a regional or access network, itself
      has a complicated topology. Receiving highly abstract routing
      information, such as a default route, would affect the sub-area's
      ability to make route selections required for traffic engineering.
      It would also limit the information passed to external ASs, for
      example, IGP-derived BGP Multi-Exit-Discriminator (MED)
      information.

      Efforts are being made to modify the IS-IS protocol to allow the
      distribution of specific route from backbone area to sub-areas. A
      mechanism facilitates such distribution is specified in [15]. When
      implementation of such mechanism become available, implementing
      multi-level IGP will be an attractive option for building routing
      hierarchy within a large network.




Yu                           Informational                     [Page 11]

RFC 2791           Scalable Routing Design Principles          July 2000


      3) One IGP Area with BGP Hierarchy

      In lieu of multi-area IS-IS, the routing hierarchy could be
      achieved by defining one IGP domain for the entire network while
      employing a BGP hierarchy. Fortunately, the hierarchical topology
      of the network in this case helps reduce adjacencies in the
      routing domain (recall there are no connections among the second-
      level network components). In addition, improvements could be made
      to further reduce the adjacency by carefully arranging the
      adjacencies to keep them at a minimum but still achieve good
      redundancy. However, this is less than ideal since the number of
      routers remains unchanged, which increases the load on the SPF
      calculation. Moreover, instability within any regional network
      would still affect the entire network (that is, there would be no
      fault isolation).

      Even with one IGP domain, it is possible to build BGP hierarchy to
      make I-BGP more scalable in the network. BGP Reflectors and BGP
      Confederations are existing mechanisms to address the scaling
      problem of full-mesh I-BGP.

      Further, a BGP reflector provides the ability to build more than
      two levels of hierarchy, as long as the interactions among the
      different levels of the hierarchy are carefully arranged to avoid
      the possibility of creating routing loops.

   Questions worth asking are: "Are two levels of routing hierarchy
   sufficient for handling scaling issues?" "Is there really a need for
   more than two levels of hierarchy?"

   When a second-tier sub-domain of a large network, such as a regional
   network, grows too big for routing protocols to handle, either
   another layer of hierarchy needs to be introduced or the sub-domain
   needs to be split into multiple second-tiered sub-domains.

   Keeping two levels of hierarchy and adding more sub-domains appears
   to be more manageable than adding another level to the hierarchy.
   However, one concern is to avoid adding more nodes to the top-level
   or transit core network to make it less scalable. Connecting the
   split sub-areas to the same core router would eliminate the need to
   add more nodes in the core area than is recommended.

   Having more than two levels of hierarchy would exceed the capability
   of IGPs as they are defined today. In OSPF, for example, all the
   areas must be connected via the backbone area, which eliminates the
   possibility of having more than two levels of hierarchy. IS-IS has
   the same limitation. Therefore, the protocols need to be redefined
   should more than two hierarchical layers in IGP be desirable.



Yu                           Informational                     [Page 12]

RFC 2791           Scalable Routing Design Principles          July 2000


   The complexity of protocols and management will increase with the
   number of levels added to the hierarchy. According to [6], most of
   the OSPF protocol bugs found over the years are related to routing
   area support. Because the interaction among the multiple levels
   increases management and debugging complexity, it is desirable to
   keep the levels within a hierarchy to a minimum.

6.2. Compartmentalization

   A scalable routing design of a large network should be able to
   localize problems or failures, thus preventing them from spreading to
   the entire network, consuming resources of network routers, and
   causing network wide instability. This is compartmentalization.
   Network compartmentalization makes fault isolation possible which
   contributes the stability of a large network.

   To achieve compartmentalization in routing design for a large
   network, one needs to avoid a design where the whole large network is
   one flat routing system or routing domain. This is the reason for the
   architecture of dividing interior and exterior routing in the global
   routing system. Within a network, it is best to divide the network
   into multiple routing domains or multiple routing areas. For example,
   in OSPF, only summary route SLAs, rather than individual area routes,
   are flooded beyond the area. When an area border router aggregates
   the routes in its sub-area, instability of any route included in the
   summary route would not cause flooding of SLAs to other areas. As a
   result, router resources in other areas would not be consumed for
   handling flooding and the SPF recalculation. In other words,
   instability within each individual area would be prevented from
   spreading to the entire routing domain.

   Since building a routing hierarchy essentially divides a big routing
   area into smaller areas or domains, it help achieve the goal of
   compartmentalization.

6.3. Making Proper Trade-offs

   When designing routing for a large network, the overall goal should
   be set with considerations of routing scalability and stability. The
   trade-offs between conflicting goals should be taken into account.
   Examples of such trade-offs are redundancy vs. scalability and
   convergence vs. stability.

   Redundancy introduces complexity and increased adjacencies to the
   network topology. Redundancy also imposes the need for as many
   alternative paths as possible for each route, which increases route





Yu                           Informational                     [Page 13]

RFC 2791           Scalable Routing Design Principles          July 2000


   processing and storage burdens. Because of these problems, it may be
   necessary to sacrifice absolute redundancy in favor of a reasonable
   level that scales better for the routing system.

   Fast convergence requires that changes in network topology be
   propagated to the network as quickly as possible. Such action
   increases routing updates and, consequently, the route processing
   burden. The burden is aggravated when a network carries full Internet
   routing information, as large networks usually do, and topology
   changes happen frequently. Route dampening may be necessary to
   achieve stability at the expense of absolute fast convergence.

6.4. Reduce Burdens of Routing Information Processing

   The tasks of reducing routing processing burdens includes: i)
   strategically place the routing intelligence within the network, ii)
   avoid carrying unnecessary routing information and iii) reduce the
   impact of route flapping.

6.4.1. Routing Intelligence Placement

   A router that executes routing policies, performs route filtering and
   dampening is said to posses routing intelligence. Routing
   intelligence is needed for a network i) to enforce the business
   agreement between network entities in the form of routing policies;
   ii) to protect the integrity of the routing information within the
   network and sometimes iii) to shield a network from instability
   happening elsewhere in the Internet.

   The more routing intelligence a router has, the more resources of the
   router are needed to perform those tasks. It is logical, then, to
   place as little routing intelligence as possible on routers that
   already are heavily burdened with other tasks.

   Usually, traffic is heavily concentrated in the core of the network.
   Because traffic aggregates from the edge of the network toward the
   core, traffic is less concentrated near the edge of the network.
   Consequently, to build a scalable routing system, it is wise to place
   routing intelligence at the edge of the network, especially in the
   networks deployed with routers that do not sufficiently decouple
   forwarding and routing. In addition, pushing routing intelligency as
   close to the edge of the network as possible also serves the purpose
   of distributing computational and configuration burdens across all
   routers.

   It is also desirable to move the heavy burden of processing routes to
   out-of-band processors, freeing more resources in network routers for
   packet forwarding and handling.



Yu                           Informational                     [Page 14]

RFC 2791           Scalable Routing Design Principles          July 2000


6.4.2. Reduce Routes and Routing Information

   As discussed in Section 4.1, a large number of routes in the system
   is one of the major culprits in route scaling problems. Therefore, it
   is best to reduce the number of routes in the system without losing
   necessary routing information.

6.4.2.1. CIDR and Route Aggregation

   CIDR as specified in [10] provides a mechanism to aggregate routes
   for efficiently utilizing IP address space as well as reducing the
   number of routes in the global routing table. CIDR offers a way to
   summarize routing information, which is one of the keys for routing
   scalability in today's Internet.

   Route aggregation would not only help global Internet scalability but
   would also contribute to scalability in local networks. The overall
   goal is to keep the routes in the backbone to a minimum.

   To achieve better aggregation within the network; that is, to reduce
   the number of routes in the network, a block of consecutive IP
   addresses should be allocated to each access or regional network so
   that when a regional network announces its routes to the transit core
   network, they can be aggregated. This way, the core and other
   regional networks would not need to know the specific prefixes of any
   particular access network. Although assignment of customer addresses
   from a provider block would have to be planned to support
   aggregation, the effort would be worthwhile.

6.4.2.2. Utilize Default Routing When Possible

   The use of a default route achieves ultimate route summarization,
   which reduces routing information to minimum. Route summarization
   also masks the instability associated with an individual route, for
   example, in the case of route flapping. It's beneficial for a network
   to utilize default routing when appropriate. For example, if a
   second-tiered regional network is a stub and there is no connected
   customer requesting full Internet routing information, the regional
   network can simply point default to its connected core network.
   However, over-summarization of routing information has the danger of
   losing routing granularity and as a result, management of network
   such as traffic engineering would be adversely affected. Therefore,
   caution needs to be exercised when using default routing.








Yu                           Informational                     [Page 15]

RFC 2791           Scalable Routing Design Principles          July 2000


6.4.2.3. Reduce Alternative Paths

   Due to the requirement of reliability, the connectivity in the
   Internet is rich, resulting in many paths toward a particular
   destination. In other words, there are many alternate paths in the
   BGP routing table towards the same destination, which consumes router
   memory and adds to the routing processing burden.

   To make routing scale, it is desirable to reduce alternate paths
   while preserving reasonable redundancy. For example, on a given
   border router (such as a NAP router), one primary path plus an
   alternate path should provide reasonable redundancy. In this case, a
   third or a fourth alternate route could be discarded for the sake of
   scaling.  This is a trade-off decision every network administrator
   needs to make based on the particular needs of her network.

6.4.3. Use Static Route at Edges

   As mentioned earlier, one of the scaling issues in large networks is
   that a single router may fan out to hundreds of customer routers. As
   a result, resource consumption will be very intensive if all the
   customer routers communicate via BGP with the edge router. Is it
   necessary for the edge router to BGP with all of its attached
   customer routers?

   At first glance, it seems necessary for a customer network in a
   different Autonomous System(AS) to exchange routing information with
   the provider network via BGP. However, this is not necessarily the
   case. When a customer network is single-homed (that is, if the sole
   network connection for a customer is via its provider network), BGP