On UCS servers with RAID configured, for improved performance, it is recommended to set the Requested Write Cache Policy on the RAID controller as Write Back instead of Write Through (the default setting). The downside of using the Write Back option is that you may lose some data if a system failure occurs before the data in the cache is written to disk. Therefore, we recommend to set the Requested Write Cache Policy on the RAID controller to Write Back Good BBU. In this mode, the controller enables Write Back caching when the Battery Backup Unit (BBU) is installed and charged. It provides a good balance between data protection and performance.

If you do not configure the virtual CPUs correctly, you may run into Non-Uniform Memory Access (NUMA) performance issues. To avoid this issue, do not configure a virtual machine from using more virtual CPUs than a single NUMA node. Otherwise, it will be scheduled across multiple NUMA nodes causing memory access degradation. Generally, this means to assign no more virtual CPUs to a virtual machine than the total number of physical cores of a single CPU socket.

When using hyperthreading virtual CPUs, note that the general CPUs utilization is 30% not 100% as threading allows for other work to be done when the main thread is stalled waiting for something. The exact numbers may be different as it depends on the workloads.