Symptom
When attempting restoration of a VMware virtual machine from disk backup in SAN environments, the restore job slows down with the events in VMware client GUI displaying “clear lazy zero” or "Allocate blocks". For restores that specify thin or auto disk provisioning, this results in the performance of restores with SAN transport mode being slower than those using NBD transport mode.
Cause
Blocks for thick provisioned disks are allocated when the disks are created; blocks for thin provisioned disks are allocated as needed during the restore (using calls to disk manager APIs). If thin provisioning is chosen for the restore, the restore operation takes longer, regardless of the original disk provisioning format. The performance impact is less if NBD transport mode is used, and if a disk is recovered directly to the ESX host rather than through the vCenter. Using thick lazy zero disk formatting places an additional load on the host, requiring the host to zero blocks before writing to disk and slowing the restore process.
Resolution
To resolve this issue, perform one of the following:
-
Restore using NBD, which may be faster depending on the network connection between the host and proxy.
-
Select SAN transport and force the disk type to Thick Eager Zero. This causes a thick eager zero disk to be created on restore. All blocks are zeroed before restoring data, the writing of restored data is quicker, and the need for negotiation between the host and proxy is eliminated.
-
Restore directly to an ESX server instead of vCenter.