Real World Scenario

One of the most common problems administrators can encounter occurs when several virtual machines without limits are deployed on a new virtualized environment. The users get accustomed to stellar performance levels early in the environment lifecycle, but as more virtual machines are deployed and start to compete for CPU cycles, the relative performance of the first virtual machines deployed will degrade. One approach to this issue is to set a reservation of approximately 10 to 20 percent of a single core's clock rate as a reservation and add approximately 20 percent to that value for a limit on the virtual machine. For example, with 3GHz CPUs in the host, each virtual machine would start with a 300 MHz reservation and a 350 MHz limit. This would ensure that the virtual machine would perform similarly on a lightly loaded ESX host, as it will when that host becomes more heavily loaded. Consider setting these values on the virtual machine that you use to create a template since these values will pass to any new virtual machines that were deployed from that template. Please note that this is only a starting point. It is possible to limit a virtual machine that really does need more CPU capabilities, and you should always actively monitor the virtual machines to determine if they are using all of the CPU you are providing them with.

If the numbers seem low, feel free to increase them as needed. More important is the concept of setting expectations for virtual machine performance.