High CPU Latency %LAT_C with no CPU contention affecting VM performance
Article ID: 385894
Updated On:
Products
Issue/Introduction
VM performance is severely affected.
When reviewing esxtop, %LAT_C is consistently high, with no or minimal %RDY and %CSTP
%LAT_C is cpu latency. It denotes the percentage of time the resource pool or world was ready to run but was not scheduled to run because of CPU resource contention.
%LAT_C can be viewed in esxtop by viewing Summary stats. (type 'f', 'i' and [return] in esxtop)
%RDY is ready wait. It denotes thepercentage of time the virtual machine or world was ready to run, but was not provided CPU resources.
This value also indicates that there the host has more vCPUs to PCPUs. Ideal value should be less than 10%.
%CSTP is co-stop. It denotes cpu contention due to waiting to be co-scheduled. A high %CSTP might indicate that a VM has more vCPUs than required. Ideal value should be less than 5%.
Environment
VMware vSphere ESXi
Cause
High CPU latency can be caused by cpu contention, cpu ready, frequency scaling (power managemen/turbo)
Resolution
Although the host and the BIOS are configured correctly with the correct power management settings, for eg. high performance, validate frequency scaling using esxtop.
- Launch esxtop
- Type 'p' to switch to the power panel
- Type 'f', 'f', and [return] to view %Aperf and %Mperf values.
- %A/MPERF column will show 100 for a core that is running on nominal frequency.
- A value higher than 100 indicates the CPU frequency has been scaled up. (Turbo is enabled)
- A value lower than 100 indicates CPU frequency has been scaled down. (Power management)
- For a performant host, %A/MPERF value should be 100 or more.
To resolve the issue,
- Validate power management in the BIOS and Power management settings on the ESXi host to ensure power management is not scaling down the CPU frequency.
- CPU turbo should scale up and increase CPU frequency.
Additional Information
Feedback
