1.2 Key Characteristics of Scalable Internetworks
1.2.2 Making the network reliable and available
A reliable and available network provides users with 24-hours-a-day, 7-days-a-week access. In a highly reliable and available network, fault tolerance and redundancy make outages and failures invisible to the end user. The high-end devices and telecommunication links that ensure this kind of performance come with a steep price tag. Network designers constantly have to balance the needs of users with the resources at hand.

When choosing between high performance and low cost at the core layer, you should opt for the best available routers and dedicated WAN links. You must design the core to be the most reliable and available layer. If a core router went down, or if a core link became unstable, routing for the entire internetwork might be adversely affected.

Core routers maintain reliability and availability by rerouting traffic in the event of a failure. Networks that can deal with failures quickly and effectively are said to be robust. To build robust networks, the Cisco IOS offers several features that enhance reliability and availability. These include support for scalable routing protocols, alternative paths, load balancing, protocol tunnels, and dial backup. The following sections describe these features.

Scalable Routing Protocols
Routers in the core of a network should converge rapidly and maintain reachability to all networks and subnetworks within an Autonomous System (AS). Simple distance-vector routing protocols, such as RIP, take too long to update and adapt to topology changes to be viable core solutions. Compatibility issues sometimes require that some areas of a network run simple distance-vector protocols such as RIP and Routing Table Maintenance Protocol (RTMP, an Apple Computer proprietary routing protocol). Whenever possible, a scalable protocol such as Open Shortest Path First (OSPF) or Enhanced Interior Gateway Routing Protocol (EIGRP) should be implemented, especially in the core layer.

Alternate Paths
Redundant links maximize network reliability and availability, but they are expensive to deploy throughout a large internetwork. Links in the core layer should always be made redundant, but other areas of a network may also need redundant telecommunication lines. If a remote site exchanges mission-critical information with the rest of the internetwork, that site would be a candidate for redundant links. To provide another dimension of reliability, an organization may even invest in redundant routers to connect to these links. A network that consists of multiple links and redundant routers will contain several paths to a given destination. If a network uses a scalable routing protocol, such as OSPF or EIGRP, its routers will maintain a map of the entire network topology. This will allow the routers to reroute traffic quickly by selecting an alternate path. In fact, EIGRP maintains a database of all alternate paths just in case the preferred route is lost.

Load Balancing
Redundant links do not necessarily remain idle until a link fails. Routers can distribute the traffic load across multiple links to the same destination. This process is called load balancing. It can be implemented using alternate paths with the same cost or metric (equal-cost load balancing), or over alternate paths with different metrics (unequal-cost load balancing). When routing IP, the Cisco IOS offers two methods of load balancing: per-packet and per-destination load balancing. If process switching is enabled, the router will alternate paths on a per-packet basis. If fast switching is enabled, only one of the alternate routes will be cached for the destination address and all packets in the packet stream bound for a specific host will take the same path. Packets bound for a different host on the same network may use an alternate route. This way, traffic is load-balanced on a per-destination basis.

Per-packet load balancing requires more CPU time than per-destination load balancing. On the plus side, per-packet load balancing allows load balancing that is proportional to the metrics of unequal paths, rather than round-robin path selection, which can help utilize bandwidth efficiently.

Tunnels
Consider an IP network with Novell NetWare running IPX at a handful of remote sites. One way to provide IPX connectivity between the remote sites is to route IPX in the core. Even if only two or three offices sparingly use NetWare, this will create additional overhead associated with routing a second routed protocol (IPX) in the core. It would also require that all routers in the data path have appropriate IOS and hardware to support IPX. For this reason, many organizations have adopted "IP only" policies at the network core because IP has become the world's dominant routed protocol.

Tunneling allows an administrator a second and more palatable option: configure a point-to-point link through the core between the two routers using IP. When this link is configured, IPX packets can be encapsulated, or packaged, inside IP packets. IPX can then traverse the core over IP links, and the core can be spared the additional burden of routing IPX. Using tunnels, administrators increase the availability of network service.

Dial Backup
Sometimes two redundant WAN links are not enough, or a single link needs to be fault-tolerant, but a full-time redundant link is too expensive. In these cases, a backup link can be configured over a dialup technology, such as ISDN, or even an ordinary analog phone line. These relatively low-bandwidth links remain idle until the primary link fails.

Dial backup can be a cost-effective insurance policy, but it is not a substitute for redundant links that can effectively double throughput by using equal-cost load balancing.