Have a cluster that was built prior to my arrival and as I get more familiar with it, I am finding some peculiarities. Three hosts with total of 96 logical processors (108 MHz) and 192 GB of memory hosting about 40 VMs. My question is regarding memory, and the consumed metric. What I am seeing on a typical VM, even some considered to carry larger loads, is that the memory usage is quite low – around or less than 25%. My understanding gained from reviewing VMware’s memory statistics is that when a VM “touches” memory, that memory remains attached to the VM and is counted under the consumed metric. The granted metric describes the VMs configuration and what is presently allocated. The active metric is what the VM actually uses. So on the typical VM, the usage at maximum is no more than 25% while the consumed & granted metrics are quite near to the VM’s configured RAM allocation. Memory metrics for the cluster are similar with active at nearly 10% of granted & consumed during production hours.
For some time now, actually prior to my arrival, the cluster resources have been over-committed in regards to memory. HA & DRS are unavailable or not relevant. In fact, we are expecting a large shipment of RAM in the coming weeks to alleviate this problem. But according to my understanding, it appears as though the cluster and VMs are in need of resource tuning and that the cluster resources can support substantially greater loads when the VMs and their resources are properly configured. My opinion is that this cluster should support at least twice this load.
Any validation of my understanding or suggestions would be greatly appreciated.
