Home > Articles > Networking

Deploying IS-IS Networks

  • Print
  • + Share This
IS-IS or Intermediate System to Intermediate System protocol is a routing protocol which has recently surged in popularity among Internet Service Providers and other large networks. The IS-IS protocol, used in conjunction with the Border Gateway Protocol (BGP), offers flexibility in configuration and scalability for large and complex networks. Here, two Cisco experts share advice on deploying IS-IS in a real network.
This chapter is from the book

This chapter is from the book

Now that you understand the IS-IS routing protocol and are familiar with the concepts of multi-area routing and routing domains, you are ready to deploy IS-IS on a network. This chapter provides information about deploying IS-IS within a network, and answers questions such as Where do you place routing domain boundaries? Why? And how do you deal with IP route summarization?

Domains and Hierarchical Network Design

One for the first questions you need to ask is "Where should I put the L1/L2 borders in a network?" In other words, how do IS-IS domains overlay onto a network's hierarchical structure? Of course, before answering this question, we need to ask yet another—what sort of hierarchy is the network built on? Figure 4-1 illustrates the two most common types.

Figure 4-1Figure 4 – 1 Two models of hierarchical networks

Traditionally, hierarchical network designs have three layers of intermediate systems: the core, distribution, and access layers. Each of these layers serves a specific purpose in the network, splitting network design into three smaller problems.

  • The network core layer is tasked with forwarding packets between the different sections of the network as quickly as possible. Core intermediate systems are normally connected through a full-mesh, partial-mesh, or some sort of ring topology to other core intermediate systems throughout the network.

  • The intermediate systems within the distribution layer act as a place to aggregate traffic and to summarize routes between the access and core layers of intermediate systems.

  • The access layer is where customers (or end users) actually connect to the network.

Many very large networks are built using this three-level hierarchy, since it tends to scale well and is well understood. There is another model used in situations where latency through intermediate systems must be reduced (generally by using fast intermediate systems and by reducing the actual number of intermediate systems a packet must pass through to make it through the network), or in smaller networks. This alternate approach is the two-tier model, which is also illustrated in Figure 4-1.

  • The core serves the same purpose in a two-layer design as it does in a three-layer design—it forwards packets between the different sections of the network as quickly as possible (or gets the packets routed to a peer network as quickly as possible).

  • The aggregation layer serves as a combined distribution and access layer: It sets policies for traffic entering the network, offers a summarization point for reachable destinations within the aggregation layer, and provides traffic aggregation.

With these two models of network hierarchy in mind, let's examine several different options for laying out domains within a network.

Putting the Network in One Routing Domain

One of the most common ways to design a large-scale IS-IS network is to place all the intermediate systems in one domain. Although this approach might not seem optimal at first glance, the following list includes some of the many reasons to design an IS-IS network this way.

  • Configuring the intermediate systems in a single domain network is much simpler. All area addresses remain the same, and there is no need to consider which interfaces are in which level or domain.

  • Designing a single domain network is much simpler. The network designer doesn't need to consider where IP address summarization will take place, for instance, or where domain borders should be.

  • The network designer does not need to consider next hop reachability for BGP, tunnels, or any other applications. Since all the endpoints are in the same domain, and all the intermediate systems within a domain share the same topology databases and reachability information, there is no need to deploy route leaking or any other mechanism to ensure the next hops are always reachable.

In fact, most of the currently deployed large-scale IS-IS networks have been designed and configured as a single large level 1 or level 2 domain and have up to 1,000 intermediate systems. Single domain networks have some disadvantages as well.

  • A flooding storm or other IS-IS problem can bring down the entire network, not just some small part of it.

  • There is no border on which to summarize IP prefixes or otherwise hide information; all LSPs must be flooded throughout the network.

Before we go on, it is important to note that most of the existing IS-IS networks were built and deployed by service providers many years before the idea of route leaking was implemented. This is relevant because the network designers could only build a single domain network if they wanted to maintain reachability and optimal routing for destinations learned through BGP. In other words, a single domain was the only real option at the time. By taking advantage of the route leaking enhancement, many networks are now able to enjoy the benefits of separating the network into multiple domains.

If you do put all the intermediate systems in a single domain, what type of domain should it be? The network can be configured as a single level 1 routing domain, a single level 2 routing domain, or a single overlaid level 1/level 2 routing domain.

Configuring all the intermediate systems within a network to run both level 1 and level 2 routing causes all reachability and topology information to be flooded throughout the network twice—once as level 1 routing and topology information, and once as level 2 routing and topology information. This result does not improve scalability, so we do not suggest running an overlaid level 1/level 2 routing domain.

It's also possible to configure all the intermediate systems in the same domain as level 1–only intermediate systems, so the entire network appears to be a single level 1 routing domain. All the reachability and topology information would be flooded once, as level 1 reachability information, rather than being flooded twice. This is the traditional way large-scale IS-IS networks were initially configured. The only problem with configuring all the intermediate systems as level 1–only intermediate systems is that if you ever decide to move to a two-level hierarchy, it can be difficult to move from a single level 1 domain to a level 1/level 2 design with a level 2 domain and level 1 domains connected to it.

Finally, you can configure all the intermediate systems to be within the level 2 routing domain—while all the intermediate systems would be in the same domain, they would be configured for only level 2 routing. Adding a second level of hierarchy would be easy with this sort of a design; since the level 2 domain is already contiguous, you can just add level 1 domains off the network or reassign nodes to them. The main problem with configuring a large network as a single level 2 domain is providing reachability to CLNS end systems. If there are any CLNS end systems within the network, there won't be a level 1 domain for them to reside in, so static CLNS routes will need to be configured to provide reachability to them.

The Pure Level 2 Core

If you have decided you don't want to place all the intermediate systems in your network into a single routing domain, then you need to decide where in the network the border between the level 2 domain and the level 1 domains is going to be. One option is to take a very structured view of the design, and divide the domains so the core of the network is contained within the level 2 routing domain. Figure 4-2 illustrates a level 2–only core.

Figure 4-2Figure 4 – 2 A level 2-only core

This configuration is possible for a network with a well-defined core, such as a true three-tier network. Each level 1 routing domain could represent a different geographical domain, or a particular site, while the layer 2 core would represent a network-wide domain backbone that all of these geographic domains attach to.

Overlapping Routing Domains

The level 1 and level 2 routing domains, however, are not necessarily going to be so cleanly separated; overlapping domains tend to be the rule rather than the exception in IS-IS network designs. Figure 4-3 illustrates a network

Figure 4-3Figure 4 – 3 The level 1/level 2 border in the core

In this network, all of the intermediate systems are part of a level 1 domain, and some are configured to run both level 1 and level 2 routing. The contiguous set of level 2 intermediate systems overlaps with the contiguous set of level 1 intermediate systems within each level 1 area. The primary issue to be wary of when working through where to put the level 1/level 2 intermediate systems in this type of design is paying attention to where IP address summarization is going to take place, and configuring it correctly.

Domain Border Considerations

When deciding where to place the level 1/level 2 border, you need to consider the following factors.

  • The physical topology of the network. If there is a natural place in which to create a pure level 2 routing domain, then you should probably consider creating one.

  • Domain partitions. Consider the result of single- and double-link failures. If a single-link failure would cause a level 2 partition, you should carefully consider how to either increase or decrease the size of the level 2 routing domain to avoid this possibility.

  • Domain sizes and flooding diameters. In Chapter 3 we noted that the primary reason for breaking a network up into smaller domains is to reduce the amount of information any given intermediate system must handle when a network topology change occurs, which also results in a reduction of the flooding scope. If your network has a number of large level 1 routing domains, and a "no intermediate system" level 2 routing domain (in other words, no L2-only nodes exist), it might be worth considering changing the location of the level 1/level 2 border to balance the flooding domains.

  • IP address summarization. You should consider the IP addressing within the network, and where it can be best summarized. You can only summarize IP addresses at domain borders.

  • Traffic flow. You should consider the paths that traffic will normally take through the network, and how domain borders will affect that traffic flow. For instance, if there is a large data center that most traffic will flow to and from, it might make sense to place this data center entirely in the level 2 routing domain.

  • + Share This
  • 🔖 Save To Your Account