Print Full Article Print This Page Download the PDF E-Mail This URL

OSPF uses cost as a metric when building routing maps. Cost can be associated with any kind of input mechanism, but most implementations associate cost with available bandwidth. This is achieved by dividing a baseline value (such as 100 million) by the available bandwidth on a specific interface. For example, dividing the baseline value of 100 million by 10 million (for a 10 Mbps Ethernet segment) produces a cost value of 10 for that interface, which is 10 times more expensive than a 100-Mbps interface (which has a cost of 1). All other factors being equal, a router will prefer the lower-cost path, which would be the 100-Mbps interface with a cost of 1 in this case.

When OSPF routers build their routing maps, the total cost for all the outbound interfaces in between the router and the known destinations is incorporated in the cost-vector algorithm. Because some networks provide asynchronous transfer rates, two endpoint systems on a link may have different costs associated with a given link, so only the outbound cost toward a destination is included in these calculations.

These concepts are illustrated in "Enterprise Routing Model", which shows three routers attached to a frame relay network labeled as Area 0 (a standalone backbone area). In this example, the three routers have T1 connections to one another, with an associated cost of 64 for each of the network interfaces. However, the routers also have 128-Kbps ISDN backup links to each other (the dotted lines), which have costs of 781.

In this design, Router B would normally send all data for Router C via the frame relay network. But if that link failed, Router B would send the data to Router D for forwarding to Router C, since the cumulative cost for that path would be 128, which is cheaper than using the dial-up backup line, which has a cost of 781. If the frame relay network collapsed entirely, Router B would have no choice but to begin using the backup line. Even if it couldn't connect with Router C directly, however, Router B may be able to send data to Router C by way of an ISDN connection to Router D (at a cumulative cost of 1,562).

When sending data to the Internet, Router C would likely choose to route the data through Router D, since it has the lowest cost (Router B has an extra Ethernet segment, with a minimal cost of 1). However, Router C may end up using Router B if the San Mateo RIP routing has been given a fixed-cost metric, which represents that the San Mateo network is available through a high-speed Ethernet link rather than providing explicit costs for all the networks at San Mateo separately. In that case, Router C would see a cost of 64 plus 64 for the two hops through Manhasset, while seeing a cost of only 64 plus 1 for the two known hops through San Mateo (rather than the actual cost of 64 plus 1 plus 64, which should be associated with the three distinct network segments). As should be obvious from this example, the use of cost as a metric works only when the values have been assigned appropriately and consistently.

OSPF Database Maintenance

The process of building and maintaining the link-state database is the most complex part of understanding OSPF and unfortunately requires a working understanding of the OSPF protocol structure and mechanisms. For this reason, a completely detailed discussion on this subject is beyond the reach of this primer. However, some of the fundamental principles of the protocol are simple. For more information on this topic, we encourage you to get one of the books available on OSPF design or to read RFC 2328.

Each OSPF node within a routing area maintains its own link-state database for all the networks, routers and interfaces associated with that area. During steady-state operations, the routers simply exchange OSPF Hello messages, which are small datagrams that advertise only that a particular router is still up and running. During synchronization operations, however, a variety of complex LSA (Link-State Advertisement) messages will be exchanged, depending on the event that occurred, the state of the database and other factors.

If an interface changes state, only a small amount of database activity is required to fully integrate the information into the area databases on all the routers within that area. If a new OSPF router is brought online, however, that router will have to discover all the routers, networks and interfaces within its area, and this process can consume a significant amount of network resources.

On broadcast and multiaccess networks, OSPF supports the use of a designated router, which lets new routers obtain complete copies of the database with minimal network impact. On point-to-point networks, however, each router has to obtain link-state data from each of the other routers independently.

This database-synchronization model represents what is perhaps the greatest challenge with running OSPF in large, complex networks, since a significant amount of time can be spent maintaining database synchronization in the face of network stability problems.

However, OSPF has many other features that make it compelling for large and complex corporate networks. But despite these advantages, OSPF is overkill for small or self-contained networks, and the use of RIP can often pay greater dividends.

Eric Hall is president of EHS Co., a network technology research and testing company in San Mateo, Calif. Send your comments on this article to him at ehall@ehsco.com.