This post is a bit off topic, but nonetheless useful when you are about to set up SharePoint farms with virtual machines. At my client we are trying to find the optimum setup for the intranet farm. The initial setup is a medium farm with 2 WFE, 1 APP and a SQL cluster (2 servers). All virtualized.
The first pageload tests indicated a very slow response times for most pages (average around 60 sec.). This was with the standard VM setup that every VM starts with. That is 2 vCPU’s and 4GB memory.
For the second test we upscaled the resources to 4 vCPU’s and 12GB. Together with extending the resources we updated the VM client tools and moved the DB server to faster SSD storage which became available after the first test. Page load time dropped to an average of 11 seconds. This had to go faster. So we asked the virtualisation team to upscale to 6 vCPUs. The result was that page load times dropped to approximately 8.2 seconds. Because this farm has to replace the current intranet farm where page loads are an average 5 to 6 seconds. This had to go faster.
So upscaling the first time resulted in a performance gain of approximately 83%. The second change from 4 vCPUs to 6 only resulted in a 27% gain. For a third time we asked the virtualisation team to upscale the VMs to 8 vCPUs.
After this question we were told by the virtualisation team that upscaling is not always a good solution, because for example you have to keep in mind the ratio between vCPUs (virtual CPU) and pCPUs (physical CPU). More CPUs and more memory not always makes a VM faster. It might result in slower VMs. Read more about vCPUs and pCPUs is these two articles: