Disclaimer: We only have two pairs of 7220s, so not directly similar to your clusters of 10!
However, on both our clusters, we see the same: even distribution of APs, and very unequal distribution of clients! This has been the case since AOS 6.x, and we are now on 8.10.13 .
We dug into it once with TAC, and the explanation is that our relatively small numbers of clients (about 7K devices peak spread between the two clusters) is only a small percentage of the max capacity of our 7220s, (~25K devices max per 7220, so ~50K for the cluster of two controllers) so they don't get load balanced. Our loads of 3k-4k clients per cluster do not hit the client load balance threshold.
Conversely, we have about 1300 APs, with each cluster having about 650 APs, and each 7220 handles 1000 APs max. And, the ~325 APs per 7220 is more than 30% of the 7220 max, which does trigger balancing.
------------------------------
Steve Bohrer
IT Infrastructure, Emerson College
------------------------------
Original Message:
Sent: Aug 01, 2024 05:36 AM
From: Guy Goodrick
Subject: Cluster client load-balancing
Hello,
1 x cluster of 10 x 9240 with standby cluster of 4 x 7240XM
1 x cluster of 3 x 9240 with standby cluster of 2 x 9240
Both clusters have a conductor and standby conductor
All running 8.10.0.13
We have just replaced all 10 7240XM's in our main active cluster with 10 9240's. IP details etc didn't change. We also added an additional 9240 to our separate active cluster of 3 9240s.
The clusters appear to have formed ok, they are showing as L2 connected. But we are seeing what appears to be some odd spread of client numbers. For example the cluster of 3 x 9240s - APs are spread evenly. But there are 5,000 clients on one box and only 60 or so on each of the other boxes. Of the 60 on each of the other boxes there are no eduroam clients (these are the bulk of our users).
I understand that clients are hashed to bucket maps and the APs hold this information. Is it possible that we need to somehow prompt the controllers to perform the process of hashing the clients again to redistribute them. Looking at the clients on the box that has 5,000 clients using 'show aaa cluster essid eduroam users' shows that all clients have been assigned that box as their UAC, and only one of the other boxes has been assigned to clients as standby UAC. It all feels a bit wonky. We also want to be sure that the other two boxes are working as expected but without any eduroam clients on them it is impossible to do.
We are planning to reload the box with 5,000 users tonight to make sure that when clients move to the other boxes everything is still working for them, but it looks like all clients will end up on just one of the other boxes so we will not be able to test whether clients on the 3rd box would function properly.
Also, on our cluster of 10 x 9240s (was 10 x 7240XMs until Tuesday) we see a similar thing - 8 of the boxes look 'normal' but 2 of them are hardly used at all for clients.
Is there any advice on what we can do here?
Guy