Defining Variance

The key to the success of load balancing in an IGRP network is variance. Variance is a user-modifiable attribute that specifies the percentage by which the performance of different links can vary, yet still be considered viable paths to the same destination. This attribute applies to the entire IGRP network rather than individual links.

The Cisco IOS defaults variance equal to one, but this can be customized according to your needs. If you establish a variance of two, for example, routes with cost metrics up to two times the cost of the best route can be accepted, up to a maximum of four routes.

Note Setting variance equal to one precludes unequal cost load balancing. Variance must be set greater than one, or there can be no variance! In such cases, load balancing is limited to just those paths whose cost metrics equal the best route.

Given that IGRP was explicitly designed to support a network with a wide diversity of transmission technologies, routes can consist of virtually any number and combination of transmission facilities. Therefore, their paths will likely have very different composite metrics. Use of IGRP's variance feature enables the IOS to automatically balance traffic loads across all feasible paths that fall within the specified variance range (refer to the section titled "Unequal-Cost Load Balancing ").

To better illustrate the concept of variance and load balancing, consider the network diagram shown in Figure 10-5.

Figure 10-5: IGRP network with multiple paths of unequal costs.

In Figure 10-5, Router A has three possible paths to the gateway router. Two of these are T1 leased lines, and the third is a 512 Kbps Frame-Relay connection.

Table 10-6 summarizes the contents of Router A's routing table. Assuming that bandwidth was the only user-defined routing metric that was modified, two routes between Routers A and the gateway are of equal cost. These two paths are T1s. The third path, the 512 kbps Frame-Relay connection, has a higher cost that reflects its lower bandwidth. The T1s enjoy almost three times the usable bandwidth of the Frame-Relay connection.

The contents of Router A's routing table, given the scenario presented in Figure 10-5, are summarized in Table 10-6. As is evident by these entries, there are three serial links interconnecting Router A and the gateway router: S0, S1, and S2. S0 and S1 are the T1 lines, and S2 is the 512 kbps Frame-Relay circuit.

Table 10-6: Router A Routing Table Contents with Three Links to the Gateway

Router Name

Destination

Next Hop

Number of Hops

Cost Metric

A

Host 192.168.130.2

Local

0

500

Host 192.168.130.9

Local

0

500

Network 192.168.125

Gateway---Port S0

2

2500

Network 192.68.253

Gateway---Port S0

2

2350

Network 192.168.125

Gateway---Port S1

2

2500

Network 192.68.253

Gateway---Port S1

2

2350

Network 192.168.125

Gateway---Port S2

2

4500

Network 192.68.253

Gateway---Port S2

2

4350

IGRP automatically balances the load across the two equal-cost T1 links. Given a default variance of 1, however, IGRP shuns the higher-cost Frame-Relay link as an undesirable route. Setting variance equal to 2, however, will qualify the Frame-Relay link for use in an unequal-cost load-balancing arrangement.

Traffic is split across the three links connecting Router A and the gateway in a ratio commensurate with their cost.

Although variance is the key attribute that enabled this unequal-cost load balancing, it is not the only attribute. Alternative paths must also pass a series of other feasibility tests before they can be used in a multipath routing arrangement. These feasibility tests are intended to maintain the stability of the network.

0 0

Post a comment