The Purpose of Using OSPF in a Multiple Area Network
This section explains what multiple area networks are and how they overcome some of the shortcomings of single area networks. Multiple areas in OSPF provide one of the main distinguishing features between the distance vector protocols and the link-state OSPF.
As you learned in Chapter 6, an OSPF area is a logical grouping of routers that are running OSPF with identical topological databases. An area is a subdivision of the greater OSPF domain, sometimes known as the autonomous system. Multiple areas prevent a large network from outgrowing its capacity to communicate the details of the network to the routing devices charged with maintaining control and connectivity throughout the network.
The division of the autonomous system into areas allows routers in each area to maintain their own topological databases. This limits the size of the topological databases, and summary and external links ensure connectivity between areas and networks outside the autonomous system.
Problems with OSPF in a Single Area
To understand the true benefits of multiple areas, consider why you might decide to create multiple areas from one area.
The following symptoms that you might observe on the network provide a clue that a single area is becoming overpowered:
The SPF algorithm is running more frequently. The larger the network, the greater the probability of a network change and, thus, a recalculation of the entire area. Each recalculation also takes longer.
The larger the area, the greater the size of the routing table. The routing table is not sent out wholesale, as in a distance vector routing protocol; however, the greater the size of the table, the longer each lookup becomes. The memory requirements on the router also increase.
The topological database increases in size and eventually becomes unmanageable for the same reasons as in the previous point. The topology table is exchanged between adjacent routers at least every 30 minutes.
As the various databases increase in size and the calculations become increasingly frequent, the CPU utilization increases while the available memory decreases. This will make the network response time sluggish (not because of congestion on the line, but because of congestion within the router itself). It can also cause congestion on the link. These can result in various additional problems, such as loss of connectivity, loss of packets, and system hangs .
To check the CPU utilization on the router, use the show processes cpu command. To check the memory utilization, issue the show memory free command.
How to Determine Area Boundaries
Although you might have an obvious need for multiple areas, the practical question is how you should implement multiple areas. There are two approaches, as follows :
To grow a single area until it becomes unmanageable
To design the network with multiple areas, which are very small, in the expectation that the networks will grow to fit comfortably into their areas
Both approaches are valid. The first approach requires less initial work and configuration. Great care should be put into the design of the network, however, because this can cause problems in the future, particularly in addressing.
In practice, many companies convert their networks to OSPF from a distance vector routing protocol when they realize that they have outgrown the existing routing protocol. This allows the planned implementation of the second approach.
The Features of Multiple Area OSPF
Now that you understand why you need to control the size of the areas, you should consider the design issues for the different areas, including the technology that underpins them and their communication (both within and between the areas).
OSPF Within an Area
One of the main strengths of OSPF is its capability to scale and to support large networks. It does so by creating areas from groups of subnets. The area is seen internally almost as if it were a small organization or entity of its own. It communicates with the other areas, exchanging routing information; this exchange is kept to a minimum, however, allowing only that which is required for connectivity. All computation is kept within the area.
In this way, a router is not overwhelmed by the entirety of the organization's network. This is crucial, because the nature of a link-state routing protocol is more CPU- and memory- intensive .
Given the hierarchical nature of the OSPF network, there are routers operating within an area, routers connecting areas, and routers connecting the organization or autonomous system to the outside world. Each of these routers has a different set of responsibilities, depending on its position and function within the OSPF hierarchical design.
The following list identifies the different OSPF routers:
Internal router Within an area, the functionality of the router is straightforward. It is responsible for maintaining a current and accurate database of every subnet within the area. It is also responsible for forwarding data to other networks by the shortest path . Flooding of routing updates is confined to the area. All interfaces on this router are within the same area. This router is the only router that can operate in a single area OSPF network, other than an Autonomous System Boundary Router (ASBR).
Backbone router The design rules for OSPF require that all the areas be connected through a single area, known as the backbone area, Area 0 , or 0.0.0.0. A router within this area is referred to as a backbone router . It can also be an internal router, an ASBR, or an Area Border Router (ABR).
ABR This router is responsible for connecting two or more areas. It holds a full topological database for each area to which it is connected and sends LSA updates between the areas. These LSA updates are summary updates of the subnets within an area. Summarization should be configured for OSPF at the area border because this is where the LSAs make use of the reduced routing updates to minimize the routing overhead on both the network and the routers.
ASBR To connect to the outside world or to any other routing protocol, you need to leave the OSPF domain. OSPF is an interior routing protocol or Interior Gateway Protocol (IGP) ; gateway is an older term for a router. The router configured for this duty is the ASBR. If any routing protocols are being redistributed to OSPF on a router, the router will become an ASBR because the other routing protocols are outside the OSPF autonomous systems. Although you can place this router anywhere in the OSPF hierarchical design, it should reside in the backbone area. Because any traffic leaving the OSPF domain is also likely to leave the router's area, it makes sense to place the ASBR in a central location that all traffic leaving its area must traverse.
This router could be configured within a single OSPF area, pointing to the outside world.
Figure 8-1 shows how the different router types are interrelated. All the routers in the backbone area, area 0, are not only performing the function of ABR, or ASBR as labeled, but are also backbone routers.
Figure 8-1. Router Definitions for OSPF
Figure 8-2 shows the connectivity and functionality of the different areas. The routers will send out routing updates and other network information through LSAs. The function or type of router will determine the LSAs that are sent.
Figure 8-2. The Different Types of OSPF Areas and LSA Propagation
Five commonly used types of link-state advertisements (LSAs) exist. Cisco uses six LSAs, which are briefly described here:
The router link LSA This LSA is generated for each area to which the router belongs. This LSA gives the link states to all other routers within an area. This LSA is flooded into an area. This is identified as a Type 1 LSA.
The network link LSA This LSA is sent out by the designated router and lists all the routers on the segment for which it is the designated router and has a neighbor relationship. The LSA is flooded to the whole area. This is identified as a Type 2 LSA.
The network summary link LSA This LSA is sent between areas and summarizes the IP networks from one area to another. It is generated by an ABR. This is identified as a Type 3 LSA.
The AS external ASBR summary link LSA This LSA is sent to a router that connects to the outside world (ASBR). It is sent from the ABR to the ASBR. The LSA contains the metric cost from the ABR to the ASBR. This is identified as a Type 4 LSA.
The external link LSA This LSA is originated by AS boundary routers and is flooded throughout the AS. Each external advertisement describes a route to a destination in another autonomous system. Default routes for the AS can also be described by AS external advertisements. This is identified as a Type 5 LSA.
The NSSA external LSA Identified as Type 7, these LSAs are created by the ASBR residing in a not so stubby area (NSSA). This LSA is similar to an autonomous system external LSA, except that this LSA is contained within the NSSA area and is not propagated into other areas, but it is converted into a Type 5 LSA by the ABR.
In the section "The ABRs and ASBR Propagation of LSAs," Figure 8-3 shows the relationships between the different LSAs. This section discusses the router and network LSAs. The LSAs concerned with communication outside an area are considered later.
Figure 8-3. The Propagation of LSAs
The Different Types of Areas
It is possible to create an OSPF network with only one area. This area is known as the backbone area or Area 0. In addition to the backbone area, which connects the other areas, OSPF networks use several other types of areas. The following are the different types of areas:
An ordinary or standard area This type of area connects to the backbone. The area is seen as an entity unto itself. Every router knows about every network in the area, and each router has the same topological database. However, the routing tables are unique from the perspective of the router and its position within the area.
A stub area This is an area that will not accept external summary routes. The LSA that is blocked is Type 5. The consequence is that the only way that a router within the stub area can see outside the autonomous system is by the use of a default route. Every router within the area can see every network within the area and the networks (summarized or not) within other areas. It is typically used in a hub-and-spoke network design.
A totally stubby area This area does not accept summary LSAs from the other areas or the external summary LSAs from outside the autonomous system. The LSAs blocked are Types 3, 4, and 5. The only way out of the totally stubby area is via a default route. A default route is indicated as the network 0.0.0.0. This type of area is particularly useful for remote sites that have few networks and limited connectivity with the rest of the network. This is a proprietary solution offered only by Cisco. Cisco recommends this solution if you have a totally Cisco shop because it keeps the topological databases and routing tables as small as possible.
An NSSA This area is used primarily to connect to ISPs, or when redistribution is required. In most respects, it is the same as the stub area. External routes are not propagated into or out of the area. It does not allow Type 4 or Type 5 LSAs. This area was designed as a special stub area for applications such as an area with a few stub networks but with a connection to a router that runs only RIP, or an area with its own connection to an Internet resource needed only by a certain division.
An NSSA is an area that is seen as a stub area but can receive external routes, which it will not propagate into the backbone area and thus the rest of the OSPF domain. Another LSA, Type 7, is created specifically for the NSSA. This LSA can be originated and communicated throughout the area, but it will not be propagated into other areas, including Area 0. If the information is to be propagated throughout the AS, it is translated into an LSA Type 5 at the NSSA ABR.
It is not always possible to design the network and determine where redistribution is to occur. RFC 1587, "The OSPF NSSA Option," deals with this subject.
The backbone area This area is often referred to as Area 0, and it connects all the other areas. It can propagate all the LSAs except for LSA Type 7, which is translated into LSA Type 5 by the ABR.
Some restrictions govern creating a stub area or a totally stubby area. Because no external routes are allowed in these areas, the following restrictions are in place:
No external routes are allowed.
No virtual links are allowed.
No redistribution is allowed.
No ASBR routers are allowed.
The area is not the backbone area.
All the routers are configured to be stub routers.
The Operation of OSPF Across Multiple Areas
As you have learned so far in this chapter, there are many pieces to the puzzle of OSPF across multiple areas. Having identified the various pieces, you need to fit them together. Then you will see how the routing protocol operates across the various areas to maintain a coherent and accurate understanding of the autonomous system.
The ABRs and ASBR Propagation of LSAs
When a router is configured as an ABR, it generates summary LSAs and floods them into the backbone area. Routes generated within an area are Type 1 or Type 2, and these are injected as Type 3 summaries into the backbone. These summaries are then injected by the other ABRs into their own areas, unless they are configured as totally stubby areas. Any Type 3 or Type 4 LSA received from the backbone are forwarded into the area by the ABR.
The backbone also forwards external routes both ways unless the ABR is a stub router, in which case they are blocked.
If a summary is received from within the area, it cannot be forwarded. Summaries received from the backbone cannot be further summarized.
The flow and propagation of LSAs within and between areas is illustrated in Figure 8-3.
Certain conditions need to be met before any LSAs can be flooded out of all interfaces. The conditions that each interface must meet before an LSA can be transmitted out of that interface are given in the following list:
The LSA was not received through the interface.
The interface is in a state of exchange or full adjacency .
The interface is not connected to a stub area (no LSA Type 5 will be flooded).
The interface is not connected to a totally stubby area (no Type 3, 4, or 5 will be propagated).
OSPF Path Selection Between Areas
The OSPF routing table that exists on a router depends on the following factors:
The position that the router has in the area and the status of the network
The type of area in which the router is located
Whether there are multiple areas in the domain
Whether there are communications outside the autonomous system
Remember the sequence of events: The router receives LSAs. It builds the topological database. Then it runs the Dijkstra algorithm, from which the shortest path is chosen and entered into the routing table. The routing table is therefore the conclusion of the decision-making process. It holds information on how that decision was made by including the metric for each link. This enables you to view the operation of the network.
Different LSAs are weighted differently in the decision-making process. It is preferable to take an internal route (within the area) to a remote network rather than to traverse multiple areas just to arrive at the same place. Not only does multiple-area traveling create unnecessary traffic, but it also can create a loop within the network.
The routing table reflects the network topology information and indicates where the remote network sits in relation to the local router.
The router will process the LSAs in this order:
The internal LSA (Type 1 and 2).
The LSAs of the AS (Type 3 and 4). If there is a route to the chosen network within the area (Type 1 or 2), this path will be kept.
The external LSAs (Type 5).
Calculating the Cost of a Path to Another Area
There are paths to networks in other areas, and then there are paths to networks in another autonomous system. The costs of these paths are calculated slightly differently.
The path to another area is calculated as the smallest cost to the ABR, added to the smallest cost to the backbone. Thus, if there were two paths from the ABR into the backbone, the shortest ( lowest -cost) path would be added to the cost of the path to the ABR.
External routes are routes passed between a router within the OSPF domain and a router in another autonomous system or routing domain. The routes discovered by OSPF in this way can have the cost of the path calculated in one of two ways:
E1 The cost of the path to the ASBR is added to the external cost to reach the next -hop router outside the AS.
E2 The external cost of the path from the ASBR is all that is considered in the calculation. This is the default configuration. This is used when there is only one router advertising the route and no path selection is required. If both an E1 and an E2 path are offered to the remote network, the E1 path will be used.
At the side of the routing table is a column indicating the source of the routing information. Typically, this is the routing protocol. In the instance of OSPF, however, it includes the LSA type that provided the path.
Table 8-2 shows the codes used in the routing table.
Now that you understand the components and operation of multiple area OSPF, you should focus on some of the design implications of creating multiple areas, as described in the next section.
Table 8-2. OSPF Routing Table Codes and Associated LSAs
Routing Table Entry
1 Router Link
This is generated by the router, listing all the links to which it is connected, their status, and their cost. It is propagated within the area.
2 Network Link
This is generated by the designated router on a multiaccess LAN to the area.
3 or 4 Summary Link (between areas)
LSA Type 3 includes the networks or subnets within an area that might have been summarized and that are sent into the backbone and between ABRs. LSA Type 4 is information sent to the ASBR from the ABR. These routes are not sent into totally stubby areas.
5 Summary Link/External Link (between autonomous systems)
O E1 or O E2
The routes in this LSA are external to the autonomous system. They can be configured to have one of two values. E1 will include the internal cost to the ASBR added to the external cost reported by the ASBR. E2 does not compute the internal costit just reports the external cost to the remote destination.
Design Considerations in Multiple Area OSPF
The major design consideration in OSPF is how to divide the areas. This is of interest because it impacts the addressing scheme for IP within the network.
An OSPF network works best with a hierarchical design, in which the movement of data from one area to another comprises only a subset of the traffic within the area itself.
It is important to remember that with all the interarea traffic disseminated by the backbone, any reduction of overhead through a solid hierarchical design and summarization is beneficial. The entire network benefits when fewer summary LSAs need to be forwarded into the backbone area. When network overhead is minimized, the network grows more easily.
With this in mind, summarization is the natural consequence. As shown in Chapter 2, "IP Addressing," summarization is not something that can be imposed on a network. It must be part of the initial network design. The addressing scheme must be devised to support the use of summarization.
In designing any network, you need to consider the resources available and to make sure that none of these resources are overwhelmed, either initially or in the future. In the creation of areas, OSPF has tried to provide the means by which the network can grow without exceeding the available resources. However, this does not remove your responsibility as the network administrator to design a network that can run efficiently within the limits of the resources available. Cisco has laid down guidelines to help in the design of stable, responsive , and flexible OSPF networks.
It is also important in any design to allow for transitions or breaks in the network. OSPF has provided a cunning device called the virtual link that allows areas disconnected from the backbone area to appear directly connected to the backbone as required.
Finally, in any network design, you must consider the traditionally tricky topology of the WAN, in particular the nonbroadcast multiaccess (NBMA) connections that fall into neither one network topology nor another.
The following sections consider all of these subjects as they pertain to multiarea OSPF design.
Capacity Planning in OSPF
Although it is possible to have more than three areas (per router) in OSPF, the Cisco Technical Assistance Center (TAC) recommends that a greater number of areas be created only after careful consideration. The results of having more areas will vary depending on the router (memory and CPU), as well as network topology and how many LSAs are generated. The recommendation is not to exceed 50 routers in an OSPF area, but again, this is a guideline and not a strict rule. Remember that OSPF is very CPU-intensive in its maintenance of the databases and in the flooding of LSAs, as well as when it calculates the routing table, a process based on LSAs.
Therefore, it is not strictly the number of routers or areas that is important, but the number of routes and the stability of the network. You must consider these issues because the number of LSAs in your network is proportional to the amount of router resources required.
With this understanding, the general rules stated by Cisco for OSPF design are that the following numbers should not be exceeded:
These are not hard and fast rules. The number of routers within an area depends on many factors; for example, a stub area with a 2500 router running over Ethernet is very different from area 0, running 7500 routers over ATM. Some of the factors that influence the number of routers per area include the following:
What type of area is it: stub, totally stub, or backbone? This determines the number of LSAs and how often and how much CPU and memory each SPF computation requires.
What level of computing power do you have in the routers within the area? The smaller routers are not designed to manage large databases and to run the SPF algorithm continually.
What kind of media do you have? The higher the bandwidth on the link, the less congestion within the router as it queues the packets for transmission.
How stable is the network? How often LSAs will be propagated because of topology changes determines the need for bandwidth, CPU, and memory resources.
If the area is running over NBMA, is the cloud fully meshed? To overcome the resources required to maintain a fully meshed network, Cisco suggests that a well-designed partial mesh over low-bandwidth links reduces the number of links and thus the amount of traffic and resources required.
If the area has external connections, is there a large number of external LSAs? If the external connections are serviced with a default link, far less memory and CPU are required than if 500 external Internet links are propagated into the network.
Do you have a hierarchical design with summarization? The greater the summarization, the smaller and fewer the LSA packets that need to be propagated.
Cisco states that, normally, a routing table with less than 500 KB could be accommodated with 2 to 4 MB RAM; large networks with greater than 500 KB might need 8 to 16 MB, or 32 to 64 MB if routes are injected from the Internet.
Further information is available on the Cisco web site at http://www.cisco.com/warp/public/104/3.html#17.0 in the OSPF Design Guide.
The following sections describe how to determine the appropriate number of neighbors to which a router should be connected, or the number of areas to which an ABR should be connected. In designing a network, elements in the network that use resources, CPU, memory, and bandwidth must be evaluated and provided for, where appropriate. Luckily, Cisco has performed extensive tests to provide clear guidelines for the design and implementation of an OSPF network.
Number of Neighbors per Router
Increasing the number of neighbors increases the resources on the router that are allocated to managing those links. More importantly if there is a designated router (DR), the router that performs the DR function might become overloaded if there are many routers on the link. It might be advisable to select the DR through manual configuration to be the router with the most available CPU and memory on the segment and to ensure that the router is not selected to be the DR for more than one link.
Number of Areas per ABR
For every area to which an ABR is connected, it will have a full topology table for that area. This could result in overloading the router before it has attempted to compute the best path. How many areas a router can support obviously depends on the caliber of the router and the size of the area. A good hierarchical designwhere the maintenance of the areas is spread over a few routersnot only shares the resources, but also builds in a level of redundancy.
One of the strengths of OSPF is the ability to scale the network. You can scale the network not only through the creation of multiple areas that limit the computation and propagation of routing updates, but also through the use of summarization. In Chapter 2, summarization was dealt with in great depth. This section builds on that knowledge and applies it to the design and implementation of multiarea OSPF.
In OSPF, two types of summarization exist:
Both have the same fundamental requirement of contiguous addressing.
OSPF is stringent in its demand for a solid hierarchical design, so much so that it has devised some commands to deal with situations that break its rules of structure.
The concept of the virtual link is explained in this section, while the commands with which to implement it are given in Chapter 9 in the section, "The area virtual-link Command."
The Virtual Link
The main dictate in OSPF is that the multiple areas must all connect directly to the backbone area. The connection to the backbone area is through an ABR, which is resident in both areas and holds a full topological database for each area.
OSPF has provided a solution for the unhappy occasion when this rule cannot be followed. The solution is called a virtual link. If the new area cannot connect directly to the backbone area, a router is configured to connect to an area that does have direct connectivity.
The configuration commands create a tunnel to the ABR in the intermediary area. From the viewpoint of OSPF, the ABR has a direct connection.
The reasons such a situation might occur are as follows:
There is no physical connection to Area 0. This might be because the organization has recently merged with another or because of a network failure.
There are two Area 0s because of a network merger. These Area 0s are connected by another area (for example, Area 5).
The area is critical to the company, and an extra link has been configured for redundancy.
Although the virtual link feature is extremely powerful, virtual links are not recommended as part of the design strategy for your network. Instead, they are a temporary solution to a connectivity problem. You must ensure that you observe the following when creating a virtual link:
Both routers must share a common area.
The areas involved cannot be stub areas.
One of the routers must be connected to Area 0.
Figure 8-4 illustrates the use of a virtual link to provide a router in Area 10 connectivity to the backbone in Area 0.
Figure 8-4. Virtual Links in a Multiple Area OSPF Network
Multiple Area OSPF Over an NBMA Network
Another design consideration is the design of the NBMA network as part of the OSPF domain. There are two main ways to approach the inclusion of an NBMA network:
The NBMA network can be defined as Area 0. The reasoning is that if the NBMA is used to connect all remote sites, all traffic will have to traverse this network. If the remote sites are made satellite areas, all traffic will have to traverse the NBMA, so it makes sense to make it the backbone area. This works well in a full-mesh environment, although it results in a large number of LSAs being flooded into the WAN and puts extra demands on the routers connecting to the NBMA network.
In a hub-and-spoke NBMA network, it makes sense to assign the hub network as Area 0 with the other remote sites and the NBMA network as other areas. This is a good design if the satellite areas are stub areas because it means that the routing informationand, thus, network overheadis kept to a minimum over the NBMA cloud. Depending on the design, the rest of the network might constitute one other area or multiple areas. This will depend on the size and growth expectations of the OSPF domain.
The configuration of a basic OSPF over an NBMA network is provided in Chapter 7.