High CPU reported on Edge VM when running NSX-T 3.2.x

Products

• VMware NSX

Issue/Introduction

• NSX-T 3.2.x
• Edge VM CPU usage is high
• Edge VM Hardware in vSphere has been manually upgraded to Hardware version higher than version 13
• Checking /var/log/core/ on the edge node, there is a core dump for the dataplane service:
  /var/log/core/core.datapathd.*.gz
• In the edge log /var/log/syslog the below FATAL error is displayed:
  

  FABRIC [nsx@6876 comp="nsx-edge" subcomp="datapathd" s2comp="intel-rte" level="FATAL"] Requested more RX queues than max available for port 0max: 8 req: 9

Environment

VMware NSX-T Data Center 3.x

Cause

• By design virtual Edges run in interrupt CPU mode while Bare Metal Edges run in polling mode.
• If the VM hardware version of an of a 3.2.x Edge is upgraded to a version higher than 13, priority control is enabled.
• When priority control is enabled, it causes the Edge VM to run in polling mode, which results in higher CPU consumption.
• To determine whether the Edge is using polling vs interrupt. 

Resolution

This is a known issue impacting NSX-T 3.2.x and does not impact NSX 4.x, which supports higher hardware versions for Edge VM.

Workaround
To workaround this issue, the Edge dataplane configuration can be changed to enable interrupt mode.

 

1. SSH to the Edge as the admin user
2. To confirm the problem condition is present:
   
   # get dataplane | find Ctrl_prio_on
   Ctrl_prio_on       : True
   
   #get dataplane | find Intr_mode_on
   Intr_mode_on       : False
3. Correct the dataplane configuration to change to interrupt CPU mode.
   
   # set dataplane interrupt-mode enabled
   # set dataplane ctrl-prio disabled
   # restart service dataplane