Routing and Backbone Label Forwarding Design
All networks, whether they span whole continents or just a group of geographic regions, present design challenges that must be addressed by the network architects. Some issues are easier to tackle than others, and certain services present unique challenges. This section reviews how USCom decided to deploy its internal and external IP routing, and also how it decided to organize its Layer 3 MPLS VPN service.
We have established that USCom operates a national backbone infrastructure that spans the continental U.S. This network must support a number of different services, including Internet access and Layer 3 VPN service. During the initial Layer 3 VPN deployment, USCom decided to deploy MPLS technology to support the architecture specified in [2547bis]. This architecture provides a network-based VPN service. It was discussed in detail in Chapter 1, "Technology Primer: Layer 3 VPN, Multicast VPNs, IPv6, and Psuedowire."
Having deployed MPLS for this service, USCom also felt that it was the right technology to support fast rerouting (FRR) capability (which you'll read about in the "Network Recovery Design for Link Failures" section). Clearly, the network will need to support even more new services in the future, so USCom's selection of MPLS as its primary technology allows the company to support existing and future service requirements.
Label Distribution Protocol (LDP) is used within the backbone to allow label switching from one edge of the USCom network to the other. However, at this point in time, only the Layer 3 VPN traffic is label-switched, leaving the Internet traffic to be forwarded by normal IP forwarding procedures. The rationale behind the decision to separate VPN forwarding from standard IP forwarding was driven primarily by the desire to continue operating the Internet network in the exact same way as before Layer 3 VPN services were introduced. This avoided any changes in configuration, monitoring, troubleshooting, or any other operational procedures that were in place for Internet traffic. In addition to this, a number of technical challenges exist if Internet traffic is label-switched, including how the existing IP tools (such as NetFlow) might behave, and how network events such as denial of service (DoS) attacks can be tracked and resolved. Chapter 5, "Global Service Provider Design Study," shows how these issues can be overcome and the USCom plan to introduce these new technologies in the future.
From an internal routing perspective, USCom runs Intermediate System–to-Intermediate System (IS-IS) as its Interior Gateway Protocol (IGP), which carries the loopback interface addresses of the PE routers (IP and VPN PE routers) as well as internal link addresses. The number of internal routes is approximately 3000. USCom does not expect to have more than 1000 routers in the IS-IS routing domain within the next two years. Hence, the IS-IS network is a flat Level 2 network that avoids having to manage the complexity of multiple levels of hierarchy.
USCom measured that the flooding activity on the existing network was perfectly reasonable. The Shortest Path First (SPF) computation time was calculated on the order of 100 ms (usually closer to 60 ms), not including the routing table updates. If at some point in the future the number of IS-IS routers has to be drastically increased because of the activation of IS-IS on various edge devices such as the ADSL or Dial access routers, USCom might consider splitting the network into multiple levels (each POP would be the Level 1 hierarchy). This would be necessary to also preserve the network convergence times. (A detailed analysis of these aspects appears in [NET-RECOV].)
Separation of Internet and Layer 3 MPLS VPN Services
From a forwarding perspective, Layer 3 VPN traffic is separated from Internet traffic, where VPN traffic is label-switched across the USCom network and Internet traffic is IP-routed/forwarded. The PE routers serving VPN and Internet customers are also separate. This is primarily because the Internet service has been deployed for a number of years and USCom wanted to deploy the new Layer 3 MPLS VPN service as a separate project, without concern that it might affect the existing customer base.
The backbone network infrastructure is addressed from the 22.214.171.124/16 block. This includes all P routers, PE routers (whether Internet or Layer 3 VPN), and any other equipment within the USCom network. The P router and core-facing interfaces on the Internet and Layer 3 VPN PE routers take their addresses from the 126.96.36.199/21 range (providing IP addresses 188.8.131.52–184.108.40.206).
The Internet PE routers and IPv4 route reflectors (RRs) take their loopback interface addresses from the 220.127.116.11/22 range (providing IP addresses 18.104.22.168–22.214.171.124).
The Layer 3 MPLS VPN PE routers and VPNv4 RRs (used for the MPLS VPN service) take their loopback interface addresses from the 126.96.36.199/22 range (providing IP addresses 188.8.131.52–184.108.40.206). This block is large enough to address 1022 devices. If the service increases above this amount, the 220.127.116.11/22 range is made available.
Each Layer 3 MPLS VPN PE router has a loopback interface configured; it is used as the source address for all Multiprotocol BGP (MP-BGP) peering sessions. Likewise, each Internet access PE router has a loopback interface assigned; it is used as the source address for all IPv4 BGP-4 peering sessions.
USCom also evaluated using one of the private IP address blocks from the [PRIVATE] range for its internal infrastructure. The use of private addresses provides some protection from the Internet because it is not a routable address space. Therefore, the internal USCom network would theoretically be hidden from the outside. However, locally attached customers could still access the network—for example, by sending traffic via a default route to USCom. Therefore, the advantages of using private address space are mitigated. Also, a future acquisition of another company might present some integration challenges, so the use of private addresses for the design was rejected.
Because the Internet PE routers and Layer 3 MPLS VPN PE routers are separate, and because USCom chose to forward only VPN traffic through label switching, forwarding separation needs to occur at the LDP level. The default behavior of the LDP protocol when executing in frame-based mode is to create and distribute label bindings for every IGP learned or local (static or connected) prefix. This is unnecessary in the USCom network because only the VPN traffic is to be label-switched, and all Internet traffic is to be routed and will never need any of the allocated label space. Therefore, only the MPLS VPN PE router loopback interface addresses (255 currently) require label bindings, because they are the only destinations to which traffic is forwarded through label switching. Example 3-1 shows how LDP filtering is achieved.
Example 3-1 Filtering Label Binding for PE Router Loopback Interfaces
no tag-switching advertise-tags tag-switching advertise-tags for ldp-pe-filter ! ip access-list standard ldp-pe-filter ! Main IP VPN PE-router loopbacks permit 18.104.22.168 0.0.3.255 ! Reserved IP VPN PE-router loopback block permit 22.214.171.124 0.0.3.255
In the future, USCom may also decide to label-switch its Internet traffic. This may be achieved by either removing the LDP filtering (the configuration of which is shown in Example 3-1) or updating the LDP filter to include the Internet PE router loopback interface addresses.
Internet Service Route Reflection Deployment
The USCom RR design for Internet service is fairly typical. It follows the network's physical topology (for loop avoidance), as shown in Figure 3-7. (Only core POPs with external peering points are shown in the figure even though the design is relevant to all Level 1 POPs.) Each Level 1 POP has two Internet RRs (the backbone P routers). All Internet PE routers peer locally and are clients of these devices. All Level 1 POP RRs are fully meshed at the BGP-4 level. The aggregation P routers are also clients of these RRs.
Figure 3-7 Placement of IPv4 Route Reflectors for Internet Service
A second level of RR hierarchy is deployed between the Level 1 and Level 2 POPs. Each Level 2 POP has two RRs (which again are the exiting backbone P routers); these are clients of their nearest Level 1 POP RRs. Every Internet PE router within a Level 2 POP is a client of the local Level 2 RRs. Each Level 3 POP Internet PE router and backbone P router peers with its nearest Level 2 POP RRs (once again following the network's physical topology).
Figure 3-8 shows how the IPv4 BGP peerings are arranged between different levels of POPs and the placement of the RRs within those POPs. Note that this figure provides the typical topology, although in some cases the Level 2 POP RRs may peer with different Level 1 POPs. (In other words, one RR peers with a different Level 1 POP than the other RR within the Level 2 POP.) This depends on the RR's geographic location in the overall topology.
Figure 3-8 IPv4 POP-to-POP BGP Route Reflection
The global IPv4 BGP table currently contains approximately 155,000 Internet routes.