Adaptive load balancing

Adaptive load balancing can be enabled per LAG to resolve traffic imbalance dynamically between LAG member ports. The following can cause traffic distribution imbalance between LAG ports:

hashing limitations in the presence of large flows
flow bias or service imbalance leading to more traffic over specific ports

Adaptive load balancing actively monitors the traffic rate of each LAG member port and identifies if an optimization can be made to distribute traffic more evenly between LAG ports. The traffic rate of each LAG port is polled at regular intervals and an optimization is only executed if the adaptive load balancing tolerance threshold is reached.

The interval for polling LAG statistics from the line cards is configurable in seconds. The system optimizes traffic distribution after two polling intervals.

The tolerance is a configurable percentage value corresponding to the difference between the most and least loaded ports in the LAG. The following formula is used to calculate the tolerance:

Tolerance = (rate of the most loaded link - rate of the least loaded link) / rate of the most loaded link * 100

Using a LAG of two ports as an example, where port A = 10 Gb/s and port B = 8 Gb/s, the difference between the most and least loaded ports in the LAG is equal to the following: (10 - 8) / 10 * 100 = 20%.

Note:

Adaptive load balancing optimization is only executed if the most loaded LAG port is beyond 10% utilization for PXC LAG ports and beyond 30% for other LAG ports.
Adaptive load balancing is not supported in combination with per-link hash, mixed-speed LAG, customized hash-weight values, per-fp-egr-queuing, per-fp-sap-instance, MC-LAG, or ESM.
Contact your Nokia representative for more information about scaling for:
- LAG ID 1 to 64 in classic CLI or max-ports 64 in MD-CLI when more than 16 ports per LAG are used
- LAG ID 65 to 800 in classic CLI or max-ports 32 in MD-CLI when more than 8 ports per LAG are used
When using port cross-connect or satellite ports, adaptive load balancing rates and the tolerance threshold do not take into account egress BUM traffic of the LAG and the adaptive load balance algorithm does not take into account link-map-profile traffic.

Example

The following configuration example shows an adaptive load balancing with 20% tolerance:

MD-CLI

*[ex:/configure lag "lag-1"]
A:admin@node-2# info
    encap-type dot1q
    mode access
    adaptive-load-balancing {
        tolerance 20
    }
    port 1/1/1 {
    }
    port 1/1/2 {
    }

classic CLI

*A:Dut-C>config>lag# info
----------------------------------------------
        mode access
        encap-type dot1q
        port 1/1/1
        port 1/1/2
        adaptive-load-balancing tolerance 20
        no shutdown
----------------------------------------------