Do not overcommit memory on the underlying ESXi host / ESXi host cluster hosting a VM with enterprise application workload with high-performance requirements. ESXi host memory overcommitment highly impacts application performance. A guest operating system and therefore an instance of an application running on top of the VM container has no visibility into ESXi host memory resource consumption. If a part of the memory assigned to the VM was swapped and an application process inside of the VM will try to use it, the process will experience an unexpected high latency while accessing swapped memory (ms instead of ns, or 100000 difference). It can lead to a process freeze or even to a BSOD of the in-guest operating system.
Ensure to configure your monitoring tools to monitor VMware Cloud on AWS. The following metrics on the VM and host level should be monitored, and an alert created if the value will be above zero (0):
There are several ways to avoid negative impacts of ESXi memory overcommitment for a VM on VMware Cloud on AWS:
For example, on a four-host cluster on VMware Cloud on AWS, using i3.metal hosts you can allocate the following amount of memory without overcommitting:
512*(4-1)*0,9 == 1382 GB, where 512 GB is total RAM on the host and 4 is the number of hosts in the cluster.