Balancing traffic on aggregation channels
Since the invention of EtherChannels, link aggregation bundles provide a convenient way to increase capacity or reliability of network transport paths. How can you ensure that the traffic is properly balanced among links in a bundle and that individual links in a bundle are operating properly?
As with many things in life, aggregation bundles are a compromise. Combining multiple physical paths allow to increase capacity without forklift upgrades, but a single flow must be pinned to a single link to prevent packet reordering. Multiple links in a bundle increase the reliability of a link, but when a member link enters an erroneous state and stops forwarding traffic, the bundle as a whole remains healthy, hiding operational problems.
To help with the management of link aggregation channels, the IP Fabric platform shows outbound and inbound balancing variance for each bundle in the network and provides state analysis of each member of the link.
Uneven balancing will block the link, despite the fact that some of the links in the bundle are underutilized. Individual flows have different bandwidth, and proper balancing scheme has to be chosen for a specific network scenario. If the link aggregation channel hash algorithm uses input parameters that are shared among many flows, the link utilization in a bundle will be uneven, causing some of the links to become overloaded. Depending on your traffic patterns, as much as possible of the varied fields of the L2, L3, and L4 header values should be used to calculate the hash to provide maximum flow diversity.
Aggregation channels with the highest variance and highest utilization deserve the most attention, and should be addressed with changing of the hashing algorithm, VLAN arrangement, or replacing of aggregation channel with a native link of higher capacity.
Outbound balancing metrics are of the highest values because that is where the balancing mechanism can be changed. High inbound balancing variance point to the problematic algorithm on the opposing side and help with troubleshooting of channel polarization.
Additionally, member state analysis point to the non-compliant or erroneous state of the individual link members, which helps to prevent aggregation channel outage ahead of time by fixing the underlying issue, and prevent operational problems by pointing to a sub-optimal path.
If you’re interested to see how the IP Fabric platform can help you to manage aggregation channels in your network, contact us for a demo or a trial.