CPU Scheduling

You may see improved performance if you ensure that the virtual machine remains scheduled when the vCPU is idle may improve performance.

To determine whether a scheduled virtual machine is providing optimal database performance:

Set monitor_control.halt_desched to false.
vCenter (the VMware management utility) should normally monitor performance. However, periodically use the vSphere Client to collect additional statistical measurements, or use esxtop or resxtop.
esxtop provides performance insight concerning CPU resource contention. Database administrators should work with their VMware administrator to interpret esxtop statistics.
Generally:
- The system is overloaded if the load average listed on the first line of the esxtop CPU Panel is greater than or equal to the number of physical processors on the system.
- The "usage percentage of physical CPUs" on the PCPU line may indicate an overloaded system.
  A reasonable upper limit in production environments is 80% usage; 90% usage should be used as an alert to the VMware administrator that the CPUs are approaching an overloaded condition. However, make decisions concerning usage levels based on the criticality of the database you are virtualizing and the desired load percentage.
- These statistics are critical:
  - %RUN – percentage of total time the processes on the VMkernel are running on the processor; a high value for %RUN does not necessarily indicate that the virtual machine is resource constrained.
  - %RDY – percentage of time the processes on the VMkernel are ready to run, but are not scheduled to a core. A process on the VMKernel in a run queue is waiting for the CPU scheduler to let it run on a PCPU. A value of greater than 10% may indicate resource contention.
  - %CSTP – percentage of time the processes on the VMkernel are stopped to allow other virtual CPUs on the virtual machine to catch up (this is the unscheduled state). A value greater than 5% typically indicates that the virtual machine's workload is not balanced.