A virtual disk can be provisioned with specific allocated space (thick disk) or with space allocated on an as-needed basis (thin disk).
Thin disks conserve space, consuming only the space that is required for data. A thin disk is created the first time data is written to the disk, so the first write has extra block allocation overhead, regardless of the transport mode. For an existing thin disk, the time required for reads and writes is slightly more than required for lazy zero thick disks; but the difference in performance can be significant for high transaction applications.
After writes to thin disks using random I/O or going to previously unallocated disk areas, backup sizes may increase, even when Changed Block Tracking (CBT) is used. Disk defragmentation may reduce the backup size.
SAN restores using thin disk provisioning can be slower than LAN restores; performance can be improved by using NBD or by setting the transport mode to SAN and forcing the disk type to thick, which uses eager zero provisioning.
When you are restoring a virtual machine that has thin disks on an NFS datastore, empty blocks on the disk are not restored. Only the actual data on the disk is restored.
Thick Disk Lazy Zero and Eager Zero Disk Provisioning
Thick disks can be provisioned using lazy zero or eager zero disk formatting.
- Lazy Zero provisioning writes zeros for each write to a new block, rather than to all blocks when the disk is created, so that initial disk creation is quicker. Lazy zero thick disks provide better performance than thin disks, but worse than eager zero thick disks.
With a lazy zeroed thick disk, a read returns only used blocks.
- Eager zero provisioning writes zeros to all allocated space when creating a virtual disk. Initial disk creation takes longer, but this format provides the best performance for writes. A read for an eager zeroed thick disk returns the entire disk, including the zeros in unused blocks.
When backing up thick disks, the proxy datastore must have at least as much free space as the maximum configured disk size for the virtual machine being backed up.
If Changed Block Tracking (CBT) is not available when a full backup is performed, all blocks are included and data in unused blocks is written as zeros. When data is restored from a backup that did not use CBT, the result is an eager zero thick disk, even if the disk that was backed up used lazy zero.
When a virtual machine disk is restored using thick disk provisioning and SAN transport, the disk uses eager zero provisioning.
For more information, see About Virtual Disk Provisioning Policies.