I ran across an article whose author claimed that on recent versions of qemu,  discard=unmap was functional on virtio disks.  So I did some testing.

 

I have not run all permutations, but have tested:

  Unraid 6.9.2

  UEFI and Seabios VMs

  Q35-5.1 and i440fx-5.1 VMs

  Windows 10 and Windows 11 with the latest virtio drivers, virtio-win-0.1.196.iso

  Ubuntu 21.4

  raw and qcow2 vdisks

  vdisks on XFS and BTRFS disks

  vdisks with and without the copy-on-write attribute set

 

In all cases, the virtio disk was functionally equivalent to a virtio-scsi disk.  On file deletion, windows unmapped blocks immediately.  Linux unmapped blocks after running fstrim.

 

https://chrisirwin.ca/posts/discard-with-kvm-2020

Edited July 21, 20215 yr by aim60

Thanks for this!

Thank you so much.

Thank you for this nugget!

Hello,

 

Doesn't work for me

 

Quote

<emulator>/usr/local/sbin/qemu</emulator>
    <disk type='file' device='disk'>
      <driver name='qemu' type='raw' cache='writeback' discard='unmap'/>
      <source file='/mnt/user/domains/Gaming/vdisk1.img'/>
      <target dev='hdc' bus='virtio'/>
      <boot order='1'/>
      <address type='pci' domain='0x0000' bus='0x03' slot='0x00' function='0x0'/>
    </disk>

 

Can you help me ?

Edited May 11, 20242 yr by thymon

Did you run the Windows Defrag and Optimize drives after making the change?

Of course...

If you have the space copy the vdisk elsewhere, then copy back using cp --sparse=always, that will confirm if the vdisk growing is the problem. 

Ok, I'll try that.
I've free space in the Array.

 

thx

does this work with qcow2?

I believe it can work with some limitations if the image was created with -o preallocation=off (sparse), but it won't work as well and as efficiently as with a raw image.

to understand better, I created the vm about a year ago, as a raw image, then converted to qcow2 for additional features, such as snapshots, now I activated discard unmap and did defragmentation but it is not working, I also tried to do cp sparse but without success, is there anything I can do?

Sorry, I don't know.