Thin Provisioning Realities and Math

vmroyale — Thu, 14 May 2009 15:27:38 GMT

I have a customer that deployed NetApp's NFS as the storage for their VI3 infrastructure. After the implementation, there was some general confusion about thin provisioning and understanding how it works in the VMware VI3 environment. In researching these issues, here is what was found:

Cloning thin provisioned disks will create thick disks.
Move/Copy (SVMotion, cold migration w/move storage option) operations will convert thin disks to thick, including NFS volume to NFS volume operations.
Disks created in vCenter and via VMware Converter are created "thin" by default.
Running defrag utilities inside a Windows virtual machine will cause the associated thin disk(s) to grow to varying degrees.
When a thin provisioned disk grows, a SCSI reservation takes place.
If performance is the primary concern for a particular virtual machine, thin provisioned disks should not be used.

Bottom Line: Without additional work and/or operational procedures, cloning, storage VMotion and even cold migrations will convert thin disks to thick disks. vSphere addresses these issues by supporting thin provisioning, but in the meantime - check out Kent's blog for a great workaround for converting thick disks to thin.

Now that the operational limitations and the realities of thin provisioned disks were understood, there also was a need to determine true disk allocation and usage.

To discover what the totals are for all allocated VMDK files, run the following command from the /vmfs/volumes directory:

find . -name '.snapshot' -prune -o -name "*-flat.vmdk" -exec ls -lh {} \;

This command will exclude the hidden NetApp snapshot directory and only return the "flat" vmdk files in the listing. The output will contain this block of information:

20G	./01234a56-bc78d901/Win2003Std32SP1/Win2003Std32SP1-flat.vmdk
20G	./0ab1cd23-45efg67h/Win2003Std32SP2/Win2003Std32SP2-flat.vmdk

Adding the sizes up will show that 40Gb of space has been allocated.

The next step is to discover what the total disk used value actually is. To do this, run the following command from the /vmfs/volumes directory:

find . -name '.snapshot' -prune -o -name "*-flat.vmdk" -exec du -sh {} \;

This command will exclude the hidden NetApp snapshot directory and only return the "flat" vmdk files in the listing. The output will appear as:

3.5G	./01234a56-bc78d901/Win2003Std32SP1/Win2003Std32SP1-flat.vmdk
3.7G	./0ab1cd23-45efg67h/Win2003Std32SP2/Win2003Std32SP2-flat.vmdk

Adding the sizes up will show that 7.2Gb is actually being used on disk. These numbers can be verified by viewing the free space value of the datastores in the VMware Infrastructure Client, NetApp FilerView or NetApp System Manager application.

Dividing the combined values returned from the "du" command by the combined values returned from the "ls" command will give the total percentage of disk space in use. In the example above, this value works out to 18% or a savings of 82%. The customer was actually seeing a savings of 51% in their production environment, and this is just by using thin provisioning. A-SIS, or deduplication, will be implemented soon, and it will be interesting to see what the disk usage numbers change to then.

Thanks for reading!

Brian Atkinson's Blog

Thin Provisioning Realities and Math