In case you are planning to install VMware ESXi on a host server, you need to be aware of the VMware ESXi best practices that get the most out of your system. You need to be able to identify the best options regarding the CPU cores and sockets, hyper threading, NUMA mode, Hardware assisted virtualization, power management, cache prefetching and memory scrubbing.
Confirm that you’re running the most recent version of the BIOS released for your system. This is crucial since you don’t want any resolved bugs or new features missing from your system.
According to Vmware ESXi best practices, to get the best performance out of your CPU, you need to check that the BIOS is set to enable all processor sockets and to enable all cores in every socket. Also, set “Turbo Boost” within the BIOS if your processors support it.
To further enhance CPU performance, confirm that hyper-threading is enabled within the BIOS if supported.
Some NUMA-capable systems offer the possibility within the BIOS to deactivate NUMA by activating node Interleaving. According to ESXi best practices, it is best to leave NUMA enabled since you get the most effective performance by deactivating node interleaving.
Check that any hardware-assisted virtualization preferences (VT-x, AMD-V, EPT, RVI, so on) are enabled in the BIOS.
Disable from inside the BIOS any devices you won’t be utilizing, like unnecessary serial, USB, or network ports.
Cache prefetching mechanisms (usually known as DPL Prefetch, Hardware Prefetcher, L2 Streaming Prefetch, or Adjacent Cache Line Prefetch) typically aid performance, particularly when access memory patterns are regular. In case you are running applications that access memory at random, however, deactivating these mechanisms may possibly lead to better performance.
VSphere best practices suggest to leave memory scrubbing rate at the default setting if the BIOS permits to be configured.
Since VSphere embraces full-featured power management capabilities, it is best to set power management within the BIOS to “OS Controlled mode”, however, there are specific cases reported where “Performance Mode” does much better.
VSphere best practices suggest to enable all deeper C-states than C1 (ie C3,C6 etc) in BIOS and use ESXi host power management to control it.
Note that C1E halt state may improve the performance of some single-threaded workloads or decrease performance for workloads that have low CPU utilization and are highly sensitive to I/O latency.