Glossary

CPU Co-Stop

CPU Co-Stop applies only to virtual machines with more than one vCPU. It is the percentage of time a VM that was ready to run stayed in a co-deschedule state, which esxtop reports as %CSTP. Broadcom's esxtop guide sets its threshold at 3 percent and advises reducing the number of vCPUs of the VM.

XorMon

How XorMon monitors it

✓XorMon graphs CPU Co-Stop as the percentage of time a vSMP virtual machine was ready to run but delayed by co-vCPU scheduling contention (100% = %RUN + %RDY + %CSTP + %WAIT).
✓The metric applies to all VMs with more than one vCPU, and it basically means that the physical processors do not have sufficient CPU scheduling opportunities.
✓The XorMon documentation suggests first checking whether many VMs or only a few big ones are affected. Big VMs can be helped by lowering their vCPU count or moving them to a host with more physical cores per processor; if almost all VMs show Co-Stop, the CPU overprovisioning factor may be too high.
✓It notes that an overprovisioning factor (allocated vCPUs divided by physical cores) of 3:1 should not cause issues, and that going above 5:1 is highly likely to have a significant impact.
✓The "CPU Co-Stop" graph of a VM is shown in milliseconds and in percent, and the data comes from the vCenter REST API.

XorMon documentation: VMware CPU monitoring: CPU Co-Stop

CPU Co-Stop

How XorMon monitors it

Related terms

Sources