Wireless Access

 View Only

  • 1.  ArubaOS 8.7.1.5 Cluster Client Balancing

    Posted Dec 20, 2021 03:08 PM
    Hello everyone,

    We've been on ArubaOS 8.7.1.5 for a few months now but just implemented clustering today. We one MM and two MDs. The APs load balance nicely between both controllers. However, the clients are mainly staying on one controller. At the moment, 478 clients on one and 9 on the other.

    Is this normal? Due to the holiday break, we don't have a lot of activity on campus at this time.

    ------------------------------
    Nathan Kuhl
    ------------------------------


  • 2.  RE: ArubaOS 8.7.1.5 Cluster Client Balancing

    Posted Dec 20, 2021 03:17 PM
    I seem to recall this being due to a MAC hashing algorithm.  I don't recall what we did to resolve it.  Hopefully someone has an answer.

    --



    Original Message


  • 3.  RE: ArubaOS 8.7.1.5 Cluster Client Balancing

    Posted Dec 20, 2021 03:23 PM
    Edited by codye Dec 20, 2021 03:27 PM
    Also depends what your rebalance thresholds are set at. Try issuing show lc-cluster group-membership on an MD to show your current threshold settings.

    https://www.arubanetworks.com/techdocs/ArubaOS_8.9.0_Web_Help/Content/arubaos-solutions/cluster/clus-load-bala.htm


  • 4.  RE: ArubaOS 8.7.1.5 Cluster Client Balancing

    Posted Dec 20, 2021 03:39 PM
    Here's the output of show lc-cluster group-membership:

    (SEM-MD01) [MDC] #show lc-cluster group-membership

    Cluster Enabled, Profile Name = "SEM"
    Redundancy Mode On
    Active Client Rebalance Threshold = 20%
    Standby Client Rebalance Threshold = 35%
    Unbalance Threshold = 5%
    Heartbeat Threshold = 900 msec
    AP Load Balancing: Enabled
    Active AP Rebalance Threshold = 20%
    Active AP Unbalance Threshold = 5%
    Active AP Rebalance AP Count = 50
    Active AP Rebalance Timer = 1 minutes
    Cluster Info Table
    ------------------
    Type IPv4 Address Priority Connection-Type STATUS
    ---- --------------- -------- --------------- ------
    self 10.40.0.11 128 N/A CONNECTED (Member)
    peer 10.40.0.13 128 L2-Connected CONNECTED (Leader)

    ------------------------------
    Nathan K
    ------------------------------

    Original Message


  • 5.  RE: ArubaOS 8.7.1.5 Cluster Client Balancing

    Posted Dec 20, 2021 03:53 PM

    Check out the link I posted above. It describes the various thresholds well. Make sure you note the part where it says:

    "For load balancing to be triggered for active clients, the active client rebalance threshold and the unbalanced threshold percentages must be met. Similarly, for the standby client, the standby client rebalance threshold and the unbalanced threshold percentages must be met."

    Remember that active client rebalance threshold =  the actual active load on a cluster member. A threshold set at X% is X% of the capacity of a platform.

    So depending on your MD platform and its client capacity, you may not be triggering load balancing with your (currently low) client count.




  • 6.  RE: ArubaOS 8.7.1.5 Cluster Client Balancing

    Posted Dec 20, 2021 05:09 PM
    Edited by nkuhl1630 Dec 20, 2021 05:44 PM
    Thanks for the reply. This just seems unnecessarily complicated, IMO. If we have 2000 clients at any given time a day, I thought the cluster would put 1000 on one controller, and 1000 on the other.

    We have two 7210 controllers. The capacity of a 7210 is 16,384 clients. At a 20% threshold, load balancing doesn't kick in until one of the controllers hits 3276 clients. Am I calculating this correctly? If so, we'll need to lower our threshold much lower.

    ------------------------------
    Nathan K
    ------------------------------

    Original Message


  • 7.  RE: ArubaOS 8.7.1.5 Cluster Client Balancing

    Posted Dec 21, 2021 06:40 AM
    In reality, clients come and go and the defaults should work fine.  It will never be perfectly be even on both controllers because again clients come and go.  It will never be perfectly be even without constantly moving clients to one controller or another, and you don't want that, because you want the traffic for a client to leave the same controller consistently.   If you recently introduced a controller into a cluster, it should gradually attempt to even out the clients

    The "complicated" method is there for customers to feel the need to tweak it and it should be left alone for customers who do not want to tweak it.  Unless the controller is at 80% capacity for clients, it should be fine.

    ------------------------------
    Any opinions expressed here are solely my own and not necessarily that of Hewlett Packard Enterprise or Aruba Networks.
    ------------------------------

    Original Message


  • 8.  RE: ArubaOS 8.7.1.5 Cluster Client Balancing

    Posted Dec 22, 2021 11:32 AM
    I get that both controllers will never be perfectly even but as I'm looking it right now, 2 days into the cluster, 499 clients in total. 476 on controller 1 and 23 on controller 2.

    Even brand new clients are getting put on 1 instead of 2. This is why I asked the following:

    ---We have two 7210 controllers. The capacity of a 7210 is 16,384 clients. At a 20% threshold, load balancing doesn't kick in until one of the controllers hits 3276 clients. Am I calculating this correctly? If so, we'll need to lower our threshold much lower.

    ------------------------------
    Nathan K
    ------------------------------

    Original Message


  • 9.  RE: ArubaOS 8.7.1.5 Cluster Client Balancing

    Posted Dec 22, 2021 11:41 AM
    Why do you need to lower the threshold?  There isn't any problem with having most of the clients on one controller given the number you have currently.  You aren't taxing the one controller.  It may seem weird, but this isn't a problem.

    Original Message


  • 10.  RE: ArubaOS 8.7.1.5 Cluster Client Balancing

    Posted Mar 07, 2025 04:00 AM

    Aruba cluster design is to operate in active/active, not active/standby. In this design it doesn't make any sense to have 90% clients on one controller and 10% on the other. Even if it is quick to fail over from the "active" controller, it makes an unnecessary higher percent risk during that failover process. 

    Hopefully this is better when we go to AOS10. 

    We have two different clusters that operate under the same settings on same code version. Yet one clusters splits clients 1700/300, the other 1300/1300. Neither of these are hitting the 20% max of 7720 yet one is operating ideal and the other isn't. The algorithm when working could have improvements, but it clearly needs some more attention during code release testing. 


    Original Message


Related Content