VMware recently released a white paper on the performance and best practices of running a Pivotal Greenplum database cluster in virtual machines. The paper reports the results of two studies. In each study, six physical machines are used in the Greenplum cluster. Three different big data workloads are run on the physical machines, and then on virtual machines in the same configuration.

One experiment compares a physical setup to a virtual configuration for running Greenplum segment servers, one per host. The response times of all the workloads in the virtual environment are within 6% of those measured in the physical environment.

Another test shows the performance impact of deploying multiple, smaller virtual machines instead of a single, large virtual machine on each segment host. The results from this test show that vSphere 5.5 provides a reduction of 11% in workload process time when compared to the same hardware configuration in a physical environment. The main performance gain occurs when each smaller virtual machine fits into a NUMA node on the physical host. For more information, please read the full paper: Greenplum Database Performance on VMware vSphere 5.5.