So according to this
http://www.codyhosterman.com/2016/12/allocation-unit-size-and-automatic-windows-in-guest-unmap-on-vm...
it looks like windows guest UNMAP isn't working right for NTFS default 4K block size, but appears to work for 32K and 64K.
Some light testing on my end (ESXi 6.5 host, guest windows 10)(copy some large files, run disk optimizer which acknowledges the disks as thin provisioned) showed similar results with a VAAI enabled array (NexentaStor in my case), with the added quirk that on a conventional server internal RAID array without VAAI, guest unmap was working at the VMFS level as well (shrinking the VMDK).
Is this a bug, or an artifact of the mode pages, or something to do with alignment on system/OS disks? Almost nobody has OS disks set at 32K/64K so it does little to help.
Considering the old UNMAP fling isn't supposed to be used for more recent Windows OS, it makes disk maintenance rather annoying.
