Database Reference
In-Depth Information
Thin Provisioned disks and Thick Provisioned Lazy Zero disks have similar
performance characteristics. This is because each time a block of data is initially
written to either a Thin or Thick Lazy Zero disk, the block must first be zeroed out. This
magnifies the write IO impact of blocks that have not previously been written because
two write IOs will be issued by the ESXi host for each block. This may have a
noticeable negative impact on write IO latency, depending on your underlying storage.
The reason to choose between Thin or Thick Lazy Zero therefore has little to do with
performance and more to do with manageability and efficiency of storage consumption.
There are tradeoffs to each choice. Your choice needs to be based on your requirements
and circumstances.
Using thin provisioning may allow for higher utilization of storage capacity as each
VMDK, data store, and underlying storage device will have a percentage of free space
unused as a safety buffer. At the same time, it will add additional management
overheads to the environment because administrators have to ensure they do not
excessively over-provision real storage resources in terms of both capacity and
performance. You need to be aware of possible growth storms and keep on top of your
storage demands.
Caution
Thin Provisioned VMDK growth operations on VMFS data stores generate
metadata updates. Each metadata update requires a lock for a brief period of time
on the VMFS data store. On some older storage arrays that do not support
VMware's API for Array Integration (VAAI) and where there is an excessive
number of Thin VMDKs or VMs per data store, this can cause SCSI reservation
conflicts, which may result in degraded performance (additional latency).
VMFS5 volumes newly created on arrays that support VAAI will use Atomic
Test and Set Locking (ATS) Only. ATS addresses the problems that used to be
caused by SCSI reservation conflicts. When selecting a storage array for use with
VMware vSphere 5.x and SQL Server, you should ensure it supports VAAI.
VMFS5 volumes that were upgraded from VMFS3 may fall back to using SCSI
reservations in certain cases. See VMware KB 1021976 and
The capacity savings from thin provisioning may well be enough to justify the
management overheads because you are able to purchase more on demand instead of up
front, and this could save a considerable amount of money. But you need to make sure
you can get the performance you need from the capacity that has been provisioned and is
used. Sizing for performance may necessitate much more capacity is provisioned on the
backend storage devices and therefore diminishes any savings that may have been had
