Hi all, we've been asked to consolidate a bit (more VMs per host) so we're experimenting to see how the environment responds to moderate memory pressure. It's been years since I've worked in a memory-constrained environment but I seem to recall ESXi ballooning rather aggressively prior to engaging in swap activity. In this 7.0 environment we're seeing hosts start to engage swap while 90-95% of the VMs have yet to experience any ballooning.
Are there any knobs we can twist to make swap more of a last resort and to look for balloon "opportunities" more aggressively? We have quite a few VMs that are generally idle and probably have more RAM than they need. One would surmise vmmemctl is operating in a target-rich environment and should have plenty of opportunities to cut fat out of dozens of VMs.
The second thing we've noticed is that after ESXi swaps memory from a VM, in many cases it doesn't seem to get returned until the VM becomes active and touches those swapped pages. Does ESXi not "swap-in" all guest pages automatically when its memory pressure has subsided?
Lastly, is there any value in tweaking Mem.MemZipMaxPct? Memory compression has always been a little disappointing in the VMware world in terms of overall memory saved and I wish our hosts would make better use of the technology.
