09-03-2024 12:56 AM
Description of the problem
In the architecture that can be seen in the figure below, we have two switches of the same type that are connected to each other with a number of links. The links in question are configured as port channels on each switch. According to the Cisco document, there is a feature called symmetric load-balancing to distribute the load between the links in a port channel. This feature works in such a way that the traffic that is balanced on a port according to the algorithm will always be balanced on the same port. The remarkable thing is that this behavior is the same in both switches. That is, when "Traffic One" on "Switch One" is balanced on the second link, it will forever pass through this link. Also, in both switches, this traffic is balanced on the same link.
The problem arises when these two switches are under load, after a period of time has passed and by performing a large number of shots and writing ports, this symmetric mode is lost, and in this case, the traffic of one that was previously on switch one It used to be on link two, now it is on link three, and the worse thing is that this traffic on switch two is placed on a link other than three, and server-to-client and client-to-server traffic sessions pass on different links. This is if both sides of the meeting used to pass through Link 2).
The point is that sometimes after shooting and writing the port channel, the problem is solved and the switches send the traffic correctly again.
09-16-2024 08:04 AM
Hello @hdihd9162 ,
read the following in the load balancing on port channels section
how many port-channels are between the two Nexus switches ? only one or more then one . With only one I don't see the impact of changing the member link the flow is using unless you refer to the fact to have some member link fully utilized.
Hope to help
Giuseppe
09-16-2024 09:16 AM
If I understand your issue, you desire Etherchannel polarization, which should be the default.
Possible causes of loss of polarization include: (1) not using exactly the same hashing algorithm on different switches, (2) packet's hash attributes change, (3) actual links being used change, (4) "rotate" command option being used, or a (5) bug.
Since you mention, loss of polarization only seems to happen after some time, I would think (1) and (4) least likely.
Discover and save your favorite ideas. Come back to expert answers, step-by-step guides, recent topics, and more.
New here? Get started with these tips. How to use Community New member guide