We performed the first set of tests with 8 vCPU VMs. We increased the number of VMs up to the maximum that successfully ran, which, in this case, was 14.
The scaling, as measured in throughput, from 1 to 14 VMs is good, going from 30,000 to 307,000 OPM
Figure 5. Scale-out performance with 8 vCPU VMs
As in all of the tests in this paper, each VM had 128 GB of memory, and the database had 140 GB on disk. This means that each VM actively used most of its assigned memory for database caching. Each host in the four-node SDDC had 512 GB of RAM for a total of 2 TB across the cloud. With 14 Oracle database VMs actively running, the memory assigned to them was 1,792 GB. While it was possible to get 16 VMs powered on for a test, they required all 2 TB of RAM on the cloud, which did not leave any room for the load-driver VM or any of the other cluster-related VMs like vCenter or NSX controllers. This is why, in the tests, we limited the 8 vCPU case to 14 VMs.