IP Routing Protocols

This section touches on the subject of IP routing in an MPLS context. The MPLS paradigm builds on existing IP routing technologies, so there are few new IP routing concepts. In-depth IP routing is outside the scope of this book. The books OSPF Network Design Solutions and IS-IS Network Design Solutions from Cisco Press offer in-depth analysis of the two interior routing protocols used the most in MPLS environments.

This section covers interior routing protocol concepts in an MPLS domain, as well as some BGP route reflector and access routing protocol ideas needed in MPLS VPN environments.

Interior Gateway Protocol

Theoretically, any IGP can be used in an MPLS environment. Link-state protocols such as Open Shortest Path First (OSPF) and Intermediate System-to-Intermediate System (IS-IS) are needed for MPLS traffic engineering (TE) applications. They are also the choice of most ISPs. Today, from a Cisco perspective, features available for OSPF and IS-IS are very similar, if not identical. Link-state technology ensures the fastest convergence that is loop-free in terms of route calculation.

From an operational point of view, link-state protocols are easier to troubleshoot because all routers have the same link-state database. In particular, the advantage of IS-IS over OSPF is that a router inserts all the prefixes it announces on a single protocol packet. Therefore, it is easier to find all the routing information announced by a particular router.

IS-IS also has scalability and reliability advantages over OSPF:

• Scalability IS-IS allows you to build larger areas than OSPF. Multiple OSPF areas might be necessary in the core to support the expected number of routers (core and PE routers) especially in the presence of unstable links. IS-IS has a different routing hierarchy than OSPF. The backbone concept still exists and has the same functionality for connecting areas. However, the backbone is implemented differently. It allows for more flexibility, particularly when the backbone must be extended. In IS-IS, the backbone is not an area, but a contiguous collection of area border routers. It should also be noted that a flat topology is desirable for traffic engineering applications.

• Fast convergence IS-IS has a faster convergence time, with features such as incremental SPF. Like OSPF, IS-IS uses Dijkstra's algorithm to compute the topology tree. However, IS-IS uses fewer packet types than OSPF to propagate routing information. Therefore, the time taken in determining how to react to a given packet decreases, speeding up the convergence process. Also, flooding (especially on broadcast media) is more optimal with IS-IS.

• Less resource usage IS-IS databases contain one link-state packet (LSP) per router in the routing domain or the area (depending on the routing hierarchy). All prefixes announced by a router (local prefixes, redistributed from other protocols) are part of the unique LSP that the router floods on the network. IS-IS has four different packet types in two forms, level 1 and level 2. The packet type depends on the router type and not on the nature of the prefix announced on the packet type. Therefore, the computation of the SPF tree is facilitated by the fact that all the routing information is on a limited number of LSPs for each router.

iBGP and Route Reflectors

In an MPLS VPN environment, iBGP sessions need to be set up among all Provider Edge (PE) routers. A full iBGP mesh is required among all PEs that need the same VPN information. Route reflectors (RRs) provide scalability and ease of management in that respect. RRs have the following advantages:

• They remove the requirement for a full mesh of iBGP sessions between all PE nodes. RRs relax this requirement by having the PEs peer with the RRs and then reflecting the routing information.

• They require fewer configurations as well as increase the network manageability by eliminating the full mesh requirement.

• They are more suitable for an iBGP network.

• They are better able to meet future service requirements.

• More RRs can be added as needed to increase performance.

• New RRs can be added to support special services.

RRs are recommended to be dedicated routers similar to PE routers, such as Cisco 72xx routers, as shown in Figure 5-5.

Figure 5-5. Route Reflectors Connected to the Core

In Figure 5-5, peering is configured between a PE to each of the four RRs. This means that each PE has four neighbor statements. If not for the RRs, you would need to configure neighbor statements to all the PEs in the network. A router, or pair of redundant routers, dedicated for route reflector functionality and not in the forwarding path allows for faster convergence by saving CPU and memory.

You can improve BGP convergence and scalability by grouping neighbors with the same update policies into peer-groups, making update calculations more efficient (by reducing the number of times the BGP table needs to be walked) and also lowering CPU and memory requirements. Increasing the input hold-queue and the maximum segment size (MSS) can also improve the BGP performance. This last case will be covered in Chapter 7, "Practical Applications of MPLS."

Access Routing Protocols

As shown in Figure 5-6, access routing defines the routing protocol between the edge LSR (PE router) and a CE router. The following VPN-aware access routing protocols are supported: static, eBGP, OSPF, RIP2, and EIGRP.

Figure 5-6. Access Routing Between the PE and CE Routers

When connecting PE to CE devices, BGP, RIPv2, and static routing use separate routing contexts for each VRF, and OSPF uses separate routing processes per VRF.

Most CPE routers are connected to the core through static routes. This is true of small VPN sites where the routing table has few routes. Using static routes would provide a significant advantage for security, but it would require manual intervention to configure the static routes.

Dynamic routing could be offered as an additional service option, allowing the customer to dynamically change his network addressing. If dynamic routing were implemented, eBGP would be the preferred routing protocol when customers are dual homing to multiple PEs or when a large number of routes is present.