Load Balancing with EIGRP

Typically, networks are configured with multiple paths to a remote network. When these paths are equal or nearly equal, it makes sense to utilize all the available paths. Unlike Layer 2 forwarding, Layer 3 forwarding has the capability to load-balance between multiple paths. That is, the router can send frames out multiple interfaces to reduce the amount of traffic sent to a single network connection. The key to this feature is that the network paths must be of equal cost (or nearly equal for some protocols like EIGRP). EIGRP uses a metric to compute the costs to a given network.

EIGRP Metric

The EIGRP metric can be based on several criteria, but EIGRP uses only two of these criteria by default:

■ Bandwidth: The smallest bandwidth between source and destination

■ Delay: The cumulative interface delay in microseconds along the path

The following criteria can be used but are not recommended because they typically result in frequent recalculation of the topology table:

■ Reliability: This value represents the worst reliability between the source and destination, based on keepalives.

■ Load: This value represents the worst load on a link between the source and destination, computed based on the packet rate and the configured bandwidth of the interface.

NOTE Although the maximum transmission unit (MTU) is exchanged in EIGRP packets between neighbor routers, MTU is not factored into the EIGRP metric calculations

Load Balancing Across Equal Paths

Equal-cost load balancing is the capability of a router to distribute traffic over all its network ports that are the same metric from the destination address. Load balancing increases the use of network segments and increases effective network bandwidth.

For IP, Cisco IOS Software applies load balancing across up to four equal-cost paths by default. With the maximum-paths maximum-path router configuration command, up to 16 equal-cost routes can be kept in the routing table. If you set the maximum-path to 1, you disable load balancing. When a packet is process switched, load balancing over equal-cost paths occurs on a per-packet basis. When packets are fast switched, load balancing over equal-cost paths occurs on a per-destination basis.

NOTE If you test load balancing, do not ping to or from routers that use fast-switching interfaces because these router-generated packets are process switched rather than fast switched and might produce confusing results.

Configuring Load Balancing Across Unequal-Cost Paths

EIGRP can also balance traffic across multiple routes that have different metrics, which is called unequal-cost load balancing. The degree to which EIGRP performs load balancing is controlled with the variance command.

The multiplier parameter for the variance command is a value from 1 to 128, used for load balancing. The default is 1, which indicates that only equal-cost load balancing is being performed. The multiplier defines the range of metric values that are accepted for load balancing by the EIGRP process.

NOTE By default, traffic is distributed proportionately among the links with unequal costs, with respect to the metric.

Example: Variance

In Figure 5-6, a variance of 2 is configured, and the range of the metric values, which are the feasible distances for Router E to get to network 172.16.0.0, is 20 to 45. This range of values determines the feasibility of a potential route.

Figure 5-6 Variance Example

Network

Neighbor

FD

AD

172.16.0.0

B

30

10

0

20

10

D

45

(config) #router eigrp 200 (config-router) #variance 2

FD = Feasible Distance AD = Advertised Distance

A route is feasible if the next router in the path is closer to the destination than to the current router and if the metric of the alternate path is within the variance. Load balancing can use only feasible paths, and the routing table includes only these paths. The two feasibility conditions are as follows:

■ The local best metric, which is the current feasible distance, must be greater than the best metric (the advertised distance) that is learned from the next router. In other words, the next router in the path must be closer to the destination than to the current router; this criterion prevents routing loops.

■ The metric of the alternate path must be less than the variance multiplied by the local best metric (the current feasible distance).

If both of these conditions are met, the route is determined to be feasible and can be added to the routing table.

In Figure 5-6, three paths to network 172.16.0.0 exist with the following metrics:

By default, the router places only path 2 (through C) in the routing table because it is the least-cost path. To load-balance over paths 1 and 2, use a variance of 2 because 20 * 2 = 40, which is greater than the metric through path 1.

In Figure 5-6, Router E uses Router C as the successor because it has the lowest feasible distance (20). With the variance 2 command applied to Router E, the path through Router B meets the criteria for load balancing. In this case, the feasible distance through Router B is less than twice the feasible distance for the successor (Router C).

Router D is not considered for load balancing with this variance because the feasible distance through Router D is greater than twice the feasible distance for the successor (Router C). In the example, however, Router D would never be a feasible successor no matter what the variance is. This decision is because the advertised distance of Router D is 25, which is greater than the Router E feasible distance of 20; therefore, to avoid a potential routing loop, Router D is not considered a feasible successor.

Was this article helpful?

0 0

Responses

  • Ute
    What routing protocol can be configured to load balance across paths with unequal metrics through th?
    2 years ago
  • wiseman
    How many destinations are using equal load balancing?
    28 days ago

Post a comment