VM|CPU Contention (% ) Metric change post upgrade of Aria Operations to 8.18.x

Products

• VCF Operations/Automation (formerly VMware Aria Suite)

Issue/Introduction

After upgrading Aria Operations to 8.18.x there is a step change in the VM|CPU Contention (%) metric across all monitored VMs

Environment

Aria Operations 8.18.x

Cause

The metrics used in the VM|CPU Contention (%) metric calculation changed in 8.18

One of the metrics used in the calculation is 'CPU | Other Wait (%)'. This metric is based on the ESXi counter '%VMWAIT'.

The counter is called 'Other Wait' as it groups a variety of wait states into a single counter.

In rare situations, the counter spikes to an incorrect value.

This is considered a bug in ESXi 7.0 U2 and above. See the KB below;

• Increased '%VMWAIT'

 

 

Resolution

This behaviour is expected. Generally, we see an increase in the metric 'CPU Contention'. Please adjust the values of any alerts to accommodate the new values for the environment.

 

Additional Information

Previously CPU contention was mapped to CPU latency in vCenter. This resulted in duplicate metrics in Aria Operations as there was another metric called CPU latency.

This changes fixed this duplicate. More importantly, it addresses the issue as CPU Latency measures more than contention. Theoretically, the CPU Latency (%) can hit 37.5% without the VM facing any contention, as the counter includes CPU frequency changes.