Figure 8 is a combination of the results presented in the previous section to allow for easy side-by-side comparison.
The test cases with 8 VMs or less show that more throughput is achieved in proportion to the number of vCPUs. In other words, the bigger the VM, the more throughput that was achieved. There are enough physical cores in all these test cases to essentially provide each database VM with plenty of CPU resources. There are 8 database VMs and 8 physical processors with 18 cores, each, in the cloud. The 18-vCPU VMs provide the best performance in all of the test cases with up to 8 VMs.
Figure 8. VMware Cloud on AWS scale-out performance with 8, 16, and 18 vCPU VMs
The test cases with more than 8 VMs (the right side of the chart) are interesting. These are the cases where resources are more scarce and where either CPU or memory constraints come into play in the performance results. At the 12 VM count, the 16 vCPU VM test case achieves the highest performance of all the tests with about 380,000 OPM, while the 18 vCPU test case falls off to around 275,000 OPM. While the efficiency of having a full socket dedicated to a single 18 vCPU VM provided the best performance with 8 VMs, the lack of any additional room for other VMs was limiting when more were added. Having more vCPUs than physical cores results in some real resource contention that in turns leads to lower overall performance.
This same type of resource contention also ends up affecting the 16 vCPU VMs once the thirteenth VM is added as the total throughput for that test case drops to 289,000 OPM. This is a little bit higher than what the twelve 18-vCPU VMs achieved, but still represents a significant drop from 380,000 OPM.