A physical CPU is basically one core in the processor’s package with support for multiple execution threads (e.g.,...
Intel hyperthreading technology) turned off. This would mean that a Xeon processor with six cores would provide the server with six physical CPUs. When threading support is enabled, the server will see each thread as a physical CPU, so if each one of those six cores enables support for two threads, the server will see 12 physical CPUs.
When a hypervisor is installed, each physical CPU is abstracted into virtual CPUs, which basically divides the available CPU cycles for each core and allows multiple VMs to "time share" a given physical processor core. By default, the hypervisor typically assigns one workload per vCPU (per core).
A rule of thumb is that a virtual CPU can support anywhere from four to eight VMs. Suppose that we use a conservative limit, such as four VMs per virtual CPU. If the server offers two, quad-core processors (eight cores total), expect the server to potentially support (8 times 4) 32 VMs, or more. If there are four quad-core processors installed (16 cores total), expect the server to provide (16 times 4) 64 VMs or more.
Remember that the more VMs that share a virtual CPU, the fewer CPU cycles are available for each VM, and the more constrained VM performance will be. If the workloads on a server need more CPU cycles, it’s better to deploy fewer VMs on a particular virtual CPU.
In addition, you can also allocate several virtual CPUs to a VM, but it is important to never exceed the number of processors in the server. For example, if the server has only one processor socket, leave only one vCPU assigned per VM. If the server has two processor sockets, assign no more than two virtual CPUs to a VM. If the server has four processor sockets, assign no more than four virtual CPUs to a VM.
Monitor virtual CPU states to improve performance
Selecting CPUs and hardware for virtualization
Tips for picking the right CPU for virtualization
Dig Deeper on Server hardware and virtualization
Related Q&A from Stephen J. Bigelow
Embedded systems and hypervisors go hand in hand. By understanding both, admins can maximize system benefits such as multiple OS support and legacy ... Continue Reading
Application load balancers and API gateways both manage network traffic, but in their own ways. Learn the differences between them and how to use ... Continue Reading
Developers don't have a lot of free time. Code reuse helps dev teams focus on the most value aspects of a project, so ensure everyone knows how to ... Continue Reading
Have a question for an expert?
Please add a title for your question
Get answers from a TechTarget expert on whatever's puzzling you.