My organisation operates some clusters consisting of three ESXi-Hosts each with VMware 6.5. Two nodes act as HA cluster with vSAN, the third one is the witness. There are only three VMs on the two hosts using the vSAN. The VCSA runs on the witness with extra storage.
vShpere shows very high VMkernel consumed memory - about 18 GB of the 32 GB of physical memory available: vSphere Monitor - Host Memory The two VMs need only 8 GB and 4 GB respectively and active memory is very low, but the overall consumed memory is above 30 GB and there frequently are annoying alarms "Host memory usage".
Most of the consumed VMkernel memory seems to be cache for the vSAN, see Understanding vSAN memory consumption in ESXi 6.0 U3, 6.5.0d, and later (2113954). In our case this gives about 17 GB per node, more than half the physical memory! Knowing the application, I am quite sure that this is unnecessarily high.
Questions:
- How can I get details about consumed VMkernel memory: How much is used by ESXi on its own, for the VMs and for the vSAN? How much of this is "active"?
- How can I configure the amount of cache to be allocated for the vSAN?