Poor hypervisor performance has many causes -- out-of-date software, insufficient storage, latency -- but server efficiency often goes overlooked.
Hypervisors have to be able to combat workload problems caused by bugs and poorly written code. As such, they have been designed to be some of the most effective and efficient software available for modern server platforms. But virtual machine performance does not rely solely on the software. Server hardware configurations play a crucial role in wringing the most data from every clock cycle and supporting better consolidation on each server.
Ask yourself these three questions to see if you can boost hypervisor performance from the server side.
Have you correctly configured the server to get the best performance from the hypervisor?
Almost all of today's servers support the major hypervisors, and many enable virtualization by default. It may be up to you, however, to
Further reading on hypervisor performance factors
Capabilities and limitations to bare-metal and hosted hypervisors
Multi-hypervisor deployment: What to expect
Have you efficiently distributed hypervisor resources and workloads on your servers?
With virtualized workloads, it is easy to forget they can put as much strain on a server as physical resources. The workloads may consume more memory or network bandwidth than a server has available. As advanced as modern servers are, you, the admin, still need to load balance workloads and resources to avoid bottlenecks and contention, which in turn, fosters better hypervisor performance.
Is your server resilient enough to support good hypervisor performance?
When you consider the factors that affect hypervisor performance, server availability may not automatically come to mind. Resilience itself does not directly determine the efficiency of the hypervisor, but maintaining server resiliency helps prevent downtime and disruption, which, in turn, affect the hypervisor. Often all you need to boost both resiliency and hypervisor performance is a software upgrade or BIOS setting change.
This was first published in April 2013