-
- Always use the para virtual SCSI controller on the guest VMs for high performance.
- Do not overcommit the VM host CPU.
- Enable hyper-threading for each CO node to improve the overall host performance by keeping the CPU pipeline busy and allowing the hypervisor to schedule the CPU clock cycle correctly.
- Ensure that the VM fits into the correct NUMA node configuration so that Greenplum can leverage this setting for its database engine.
- Change per device queue length that is “No of outstanding IOs” to 256 on each CO node to improve the IO concurrency of each data store.