Database Reference
In-Depth Information
Figure 5.8
Advanced setting numa.vcpu.maxPerVirtualNode.
vNUMA is set upon the first boot of the virtual machine, and by default does not change
unless the vCPU count is modified. One of the reasons this does not change is that not
all operating systems (or even all applications) tolerate a change to an underlying
physical NUMA infrastructure topology. In addition, sometimes the operating system
can adjust, but the application cannot. Therefore, make sure to understand the
applications you are working with before making changes that could negatively affect
their performance. We will discuss advanced settings later that change the default
behaviors. Before we get there, we will discuss the defaults first, because it is
important to understand what vSphere is doing under the covers and to use this
information to determine whether a change is necessary.
The method by which the vNUMA topology is set is as follows: Upon the first boot of a
vNUMA-enabled virtual machine, a check is made to see if the Cores per Socket
(virtual socket) setting has been changed from the default value of 1. If the default value
of 1 is present, then the underlying physical server's NUMA architecture is used. If the
default value of Cores per Socket has been modified, this determines the virtual
machine's NUMA architecture. See
Figure 5.9
for a screenshot of where this setting is
located.
