Thin or thick provisioning for virtualization

[DEMONmachine]

I'm using oVirt and VMWare virtualization platforms with FreeNAS 11.3 storage over NFS.
And considering about using thin/thick disk provisioning.
For virt. platform thick (preallocated) provision is usually better in terms of performance, and thih is better for space saving.
But FN (by default) is using compression.

Options:
1) thih provision + no compression. You losing performance. And overcommitting on virt. platform level.
2) thin + compression. Bad peformance and "double" overcommiting.
3) thick + compression. Better performance, overcommiting only on the storage level.

I suppose "real" storage usage for cases #2 and #3 will be almost the same, and probably a bit better than #1

So if I understanding it right there is to reason to use thin provision at all, and the best option is #3?
Are there any downsides of this option?

 

[blanchet]

I cannot tell for oVirt but for VMware the main bottlenecks are:

• Synchronous Write Performance (a ZFS SLOG helps a lot)
• IOPS (more vdevs help)

In comparison, ZFS Compression and thin vdisks have a very limited impact, so, at least for VMware, you can use compression+thin provisionning to save space without worrying to much about performance penalties. By default: FreeNAS uses compression, and VMware uses thin provisioning.