notes: EIGRP Variance and Load-sharing

The degree to which EIGRP performs load balancing is controlled with the variance command. You set the variance to a number from 1 to 128. The default is 1, which indicates equal-cost load balancing. The multiplier defines the range of metric values that are accepted for load balancing by the EIGRP process. For example, if you want load balancing to occur between two links, and one has a metric of 1000 and the other has a metric of 2000, you need to set the variance to 2 to cause load balancing between the two links. A route must be in the topology table by meeting
the initial feasibility test; otherwise, it won’t be used in load balancing regardless of the variance.

- The variance command is used to determine which routes are feasible for unequal-cost load sharing.
- Variance defines a multiplier by which a metric may differ, or vary, from the metric of the lowest-cost route.
- Any route whose metric exceeds the metric of the lowest-cost route, multiplied by the variance, will not be considered a feasible route.
- The default variance is one, meaning that the metrics of multiple routes must be equal, to load balance.
- Variance must be specified in whole numbers.
- Load sharing is per destination if the packet is fast switched or CEF switched using the default CEF configuration.
- Load sharing is per packet if process switching is used or if the CEF configuration was modified.
- CEF and fast switching can be turned off with "no ip cef" and "no ip route-cache", and then the router will perform unequal-cost, per packet load balancing.
- For a good load-sharing example refer to : http://blog.ru.co.za/2009/04/02/eigrp-metric-manipulation/
- Load-sharing by default is based inversely to the traffic-share rate among the multiple paths.
> So if wanting a traffic-share rate of 1:5, the first path would get five times more traffic than the second.
- This can be changed to use only the best metric path, even though both routes are in the table with traffic-share min across-interfaces

1. Disables CEF under the interface.

no ip cef

2. Disables fast-switching under the interface

no ip route-cache

Note:both above is necessary for per packet load-balancing via process switching)

3. Meaning a interface metric can be up to 5 times more than the current FD (def=1
Changes the default (4 paths) over which EIGRP can be set load balance.
Default: share inversely proportional to metric
Only use the best metric path, even though multiple in routing table

router eigrp 15
variance 5
maximum-paths 2
traffic-share balanced
traffic-share min across-interfaces

note: traffic-share min (not a default command)

Load Balancing

- note equal cost vs unequal cost

- traffic-share command
balanced vs. unbalanced (default)

- load-balancing
- ip load-sharing per-packet
- ip load-sharing per-destination (default)

interface FastEthernet0/0.67
no ip route-cache
ip load-sharing per-packet

unequal cost load balancing default is 4 most 6 nowadays 16
using variance -

i.e.
5 routes with the ff metric

1000
2000
3000
4000
5000

load balance 2 paths= solution variance of 2


3 paths: variance of 3


if the metrics are not multiple load balance between 3 paths.

1000
2000
2500
3000
variance 3 and maximum-path of 3 (combine multiple commands to achieve the requirements;)


load balance between 3 paths but not allowed using variance: for the ff metrics;
1000
2000
3000
4000
5000

- using offset list
- bandwidth and delay value - more pain