Reliable Memory is a unique feature which was introduced in vSphere 5.5. If the server platform supports the reliable memory feature, ESXi can put the memory pages consumed by critical services on more reliable memory regions. The memory scheduler will do its best at putting all critical code and data in reliable memory. However, if at any point of time the system is short in reliable memory, it has to fall back to allocating regular memory. At that point, using the reliable memory for critical services is a best effort.
Note: ESXi tries to remap any pages from and to reliable memory to recover from a reliable memory shortage. However, not all pages are eligible for remapping so ESXi might not be able to put back all critical services into reliable memory.
Additionally, If the amount of reliable memory on a system is too small to contain all the critical services at boot time, the host might hang or PSOD.
As a minimum, we recommend booting ESXi with 3GB of reliable memory. In case of a system with multiple NUMA nodes, we also recommend evenly distributing the reliable memory between NUMA nodes if possible.
To maintain the guarantee that all the critical services remain in reliable memory, it is recommended not to exhaust the reliable memory. In other words, configuring virtual machines with more reliable memory than the host capacity is not recommended.
Note: This issue cannot occur if a machine is configured without reliable memory. If reliability is not a concern or not a desired feature, re-configuring the machine without reliable memory will solve any associated issue.