Wireless Access

I am testing dual links and LACP on some 335s (soon 5xx's) and experiencing some weird issues. (I'm not trying to set up and use GRE striping with the controllers. Just dual links on the switch side via a port channel)

ISSUE #1:
I have an AP (AP-1 below) that had been working fine. Port channel set up properly on the cisco switch, Po saw both member interfaces, and the output of "show ap debug lacp ap-name" was showing two UAC tunnels (one to each of my controllers in my cluster). After a simple reboot of the AP, the AP came up, both interfaces on the switch are up, the Po looks good, the AP reports the two UACs using both ethernet ports. However, all outgoing UAC traffic is going out eth0 only (see red arrow).  The AP-2 example on the right shows another AP in the same builiding, with the same configuration, and it shows the UAC traffic being split out both eth0 and eth1 properly. 

Any ideas on what I can check? Again, AP-1 had been working properly, and all I did was reboot it. Both APs are in the same AP group, both are on the same cisco switch, all the switch ports have the same config (each AP has its own Portchannel defined, with its two members assigned). Everything looks fine on the cisco side too.

ISSUE#2
Looking at AP-2 above, the UAC's are spread over the two interfaces, which is good. I can see in the "GRE Traffic Sent on Enet Ports" section the TX COUNT traffic is going out each interface. However, under the "GRE traffic received on enet ports" section I see that traffic from both of my controllers is all RX on the eth0 interface. My load balance algorithm on the switch is set to source-IP, is that not good enough to split the inbound controller traffic between the two switchports in the port channel? 
​

Hello Cody, a Links Aggregation LAG (an Etherchannel to use the Cisco jargon) controls outgoing traffic only (in other terms it has no control in incoming traffic) and to egressing packets traversing a particular logical LAG interface it applies a "routing" governed by the deterministic results of the hashing algorithm ruling the LAG of the particular device you're considering (Cisco Switch on one end, Aruba AP on the other), the output of this "calculation" will determine what physical interface of the LAG will be used for a particular outgoing flow...this will potentially led to outgoing traffic "polarization" over a particular physical interface and this will happen if the hashing algorithm calculation results vary very little (say the Source/Destination IP digested by that hashing algorithm don't change...the result doesn't change too and so the selected outgoing physical port does, you will see one port used more than the other instead of seeing the pair equally used). This is valid on both ends (so from the PoV of the Aruba AP looking its peer device and also from the PoV of the Cisco Switch looking its peer device).

------------------------------
Davide Poletto
------------------------------

Original Message:
Sent: Oct 25, 2021 03:48 PM
From: Cody Ensanian
Subject: AP-335 and LACP

I am testing dual links and LACP on some 335s (soon 5xx's) and experiencing some weird issues. (I'm not trying to set up and use GRE striping with the controllers. Just dual links on the switch side via a port channel)

ISSUE #1:
I have an AP (AP-1 below) that had been working fine. Port channel set up properly on the cisco switch, Po saw both member interfaces, and the output of "show ap debug lacp ap-name" was showing two UAC tunnels (one to each of my controllers in my cluster). After a simple reboot of the AP, the AP came up, both interfaces on the switch are up, the Po looks good, the AP reports the two UACs using both ethernet ports. However, all outgoing UAC traffic is going out eth0 only (see red arrow).  The AP-2 example on the right shows another AP in the same builiding, with the same configuration, and it shows the UAC traffic being split out both eth0 and eth1 properly. 

Any ideas on what I can check? Again, AP-1 had been working properly, and all I did was reboot it. Both APs are in the same AP group, both are on the same cisco switch, all the switch ports have the same config (each AP has its own Portchannel defined, with its two members assigned). Everything looks fine on the cisco side too.

ISSUE#2
Looking at AP-2 above, the UAC's are spread over the two interfaces, which is good. I can see in the "GRE Traffic Sent on Enet Ports" section the TX COUNT traffic is going out each interface. However, under the "GRE traffic received on enet ports" section I see that traffic from both of my controllers is all RX on the eth0 interface. My load balance algorithm on the switch is set to source-IP, is that not good enough to split the inbound controller traffic between the two switchports in the port channel? 
​

------------------------------
Cody Ensanian
------------------------------