Home > Storage > PowerMax and VMAX > Storage Admin > Dell PowerMax and VMware vSphere Configuration Guide > Virtual disk allocation schemes
In the past, Dell has recommended using zeroedthick instead of thin virtual disks. Thin on thin was not recommended because using thin provisioning on two separate layers (host and array) increases the risk of out-of-space conditions for virtual machines. The PowerMax integrates these layers better, providing several options for monitoring and reacting to out-of-space issues. As a result, using thin on thin is a perfectly acceptable option for these software versions. Remember and consider that certain risks remain in doing so. Dell maintains zeroedthick VMDKs, which is the VMware default, as the best practice to avoid issues.
Despite the general best practice, Dell supports all virtual disk types, and depending on the customer situation, zeroedthick disks may not be appropriate. The VMware administrator must evaluate the requirements of the application being hosted and the needs of the application users/stakeholders when selecting a virtual disk type. Choosing a virtual disk allocation mechanism is typically based on the answers to these three questions:
While thin virtual disks are logically more efficient than zeroedthick virtual disks (less space on the VMFS), they are not physically more efficient. The thin virtual disks still use the same amount of storage as zeroedthick on the array. In other words, they are equal. As previously discussed, zeroedthick only writes data that the Guest OS writes. It does not consume more space on the device than a similarly sized thin virtual disk. The primary difference between zeroedthick and thin virtual disks is not a question of space. It is a matter of performance and the quantity or host-observed capacity of the array thin devices that are presented to the ESXi server. Due to the architecture of the array, whether multiple, smaller devices are presented to the environment, or a single large device, the consumed disk space is essentially the same. Therefore, packing more virtual machines into a single thin device is not necessarily more space efficient, but does allow for higher virtual machine density.
Remember that high virtual machine density can cause performance problems. Besides the typical issues such as device queuing, metadata locks can become a problem with thin virtual disks. Fortunately, this problem was solved with the hardware-assisted locking primitive offered by VAAI, further promoting the viability of larger devices and thin on thin.
Hardware-assisted locking provides a more granular means to protect the VMFS metadata during updates than the SCSI reservation mechanism used before hardware-assisted locking was available.
Hardware-assisted locking leverages a storage array atomic test and set capability to enable a fine-grained block-level locking mechanism. With or without VAAI, VMFS metadata updates are required when creating, deleting, or changing file metadata (among many other operations). An example of such an operation is the growth of a thin virtual disk.
VMware and Dell have ameliorated the drawbacks of using thin virtual disks on the array through the following:
These improvements can make thin on thin a more attractive choice for a default VMDK disk type in customer environments, and mostly with the PowerMax. The choice depends on how risk-averse an application is and the importance or priority of an application running in a virtual machine. Thin on thin may be used:
If thin on thin is used, you must configure alerts on the vCenter and array level.
While most of the information in the previous section applies to the PowerMax array, there are some noteworthy differences which impact which virtual disk type should be chosen. As explained in the section Eagerzeroedthick and VAAI when an eagerzeroedthick disk is created on the PowerMax, space is not preallocated on the device. The architecture of the array makes it unnecessary.
The PowerMax does not reserve space. It seems logical to conclude that there is little difference between using an eagerzeroedthick and a zeroedthick virtual disk on that platform. This determination is true from a storage performance perspective. However, if VMware or an application requires the use of an eagerzeroedthick virtual disk, that disk type must be selected when creating the virtual machine. Examples include Fault Tolerant VM and Oracle RAC.
Note: The PowerMax has mostly negated the impact of provisioning a new extent on the array. If a VMDK is loaded with data repeatedly or once and retained, it might be beneficial to use eagerzeroedthick and even persistently allocate the device on the PowerMax. Doing so reserves and preallocates the extents on the device so that there is no overhead (as small as it might be) in loading data into the datastore. This configuration could be important if the data load is time sensitive and the amount of data significant.