Thin provisioned are sparse files. Meaning only the actual used data is stored, everything that's still empty is skipped. In order to write the system needs to make room for it. Normally this isn't much of an issue but it does require a bit more I/O because stuff needs to be moved around. With I/O intensive applications all these little bits of extra I/O add up costing you performance.
With "thick" provisioning the whole size of the disk is created. Even without any data in it it would be the exact size (filled with zeros). Now when you write to it it can do so immediately (the space is already there).
But note that I/O on a guest will always be comparatively slow (even if you use virtio). So depending on the load of your database it may not be a good idea to virtualize it. If you need the absolute maximum I/O performance run it on real iron, not a VM.
As far as ZFS is concerned I do not believe there is any difference between standard or sparse zvols, other that the space being reserved for a standard zvol. In fact you can switch between them by changing the relevant reservation property.
The only real pro for a "thick" zvol is that it makes sure enough space is always available for the volume.
Remember that it is the refreservation that you likely want. What sort of pool is this on? RAIDZs are extremely poor (for efficiency and performance) for storing ZVOLs with small blocks on...
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.
