Wireless Access

We have a cluster of 2 7010 controllers and a MM running 8.5.0.1.  We regularly see severely unbalanced distribution of clients between cluster members; ie, right now we have 89 clients on one controller, and 391 on another.  This is typical since we upgraded from 6.5.  We had a TAC case open (no help so far) and have performed troublehsooting steps as per forum posts (ie, https://community.arubanetworks.com/t5/Wireless-Access/AOS-8-3-Cluster-no-client-balancing-and-crashes/td-p/476665).  The cluster looks healthy, is L2 connected with very low missed heartbeats.  Should we be seeing a better distribution between controllers?

What is the total number of clients?

The 7010 supports 2048 users. Client loadbalancing occurs only if certain thresholds are met.

What is the output of " show lc-cluster group-profile <name of the cluster profile>" ?

--Give Kudos: found something helpful, important, or cool? Click Kudos Star in a post.
--Problem Solved? Click "Accepted Solution" in a post.

You say that certain thresholds need to be met before balancing happens - what are those thresholds?

The thresholds applicable to client load balancing are the following

Active client rebalance threshold

standby client rebalance threshold

Unbalanced threshold

The link below gives an idea as to what these thresholds are,

https://www.arubanetworks.com/techdocs/ArubaOS_83x_Web_Help/Content/ArubaFrameStyles/Cluster/Load%20Balancing.htm

These parameters can be checked using the command, " show lc-cluster group-membership"

Now in your case, you have two 7010 controllers whose platform limits are 2048 devices per controller.

For active clients to be successfully loadbalanced they need to meet two thresholds - The Active client rebalance threshold and the unbalanced threshold.

So again, could you post the output of " show lc-cluster group-membership " so as to verify what the thresholds are set for and that they are met.

--Give Kudos: found something helpful, important, or cool? Click Kudos Star in a post.
--Problem Solved? Click "Accepted Solution" in a post.

Here's the output of that command:

([REDACTED]_MC2) *#show lc-cluster group-membership

Cluster Enabled, Profile Name = "[REDACTED]_Cluster"
Redundancy Mode On
Active Client Rebalance Threshold = 50%
Standby Client Rebalance Threshold = 75%
Unbalance Threshold = 5%
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
---- --------------- -------- --------------- ------
peer 10.1.100.17 128 L2-Connected CONNECTED (Leader, last HBT_RSP 1ms ago, RTD = 0.000 ms)
self 10.1.100.16 128 N/A CONNECTED (Member)

From the output, the Active Client Rebalance Threshold is 50%.

For a 7010 controller this equates to 50% of total capacity which is 50% of 2048.

This results in 1024 active clients, this is the number of clients that need to be active on one of your controllers to meet condition one i.e., Active client rebalance threshold.

Now as for Condition 2,  Unbalance Threshold is 5%

This put in a equation looks like this,

[(Active clients / MD-1 total capacity) - (Active Clients / MD-2 total capacity)*100] > 5%

For example,

Lets say MD-1 (7010) has 1200 clients and MD-2 (7010) has 950 clients.

Active client rebalance threshold is met on MD-1

Unbalance threshold : [ (1200 / 2048) - (950 / 2048) * 100] = 12% > 5 %

Since the unbalance threshold is also met, the clients will be load balanced in a hitless manner without disrupting the services.

This is more or less how client LB is triggered.

Note : The equations are for your understanding only

However these knobs can be altered based on your total client count to meet the client load balancing thresholds (Just for the aesthetics of it).

What is your total client count?

--Give Kudos: found something helpful, important, or cool? Click Kudos Star in a post.
--Problem Solved? Click "Accepted Solution" in a post.

This is great, thanks for the explanation!  We typically see 2500 concurrent clients at peak.