Your opinion of Thin Provisioning, Fragmentation and SCSI Reservations
Dear community members,
this is about Thin Provisioning (ESX, not Storage), Fragmentation and SCSI Reservations.
From what I understand after reading VMware's White Paper about Virtual Disk Thin Provisioning, fragmentation and SCSI reservations are nothing what really impacts the system significantly.
However, after reading this and a lot of other documentations as well as trying things in my lab environment at work, I do have some thoughts/questions that I would like to share and hear opinions of others.
1.) Initial VMDK size and incremental growth size
When creating a thin provisioned disk, it initially does not consume any space on the VMFS and grows as needed in increments of the VMFS's block size (1 MB by default). Wouldn't it make sense - in terms of less defragmentation and SCSI reservations - to be able to set an initial (minimum) size of the virtual disk as well as the size of the increments. (E.g. for a Windows file server set the initial size to 5 GB and the growth increments to 100 MB.)
This way, defragmentation and SCSI reservations could completely be avoided for the initial installation of the OS and significantly reduced during normal operation.
As of now, no defragmentation tool is available. The ways to defragment a thin provisioned disk are Storage vMotion, Migration...
Assuming I have a VMFS with a lot of thin provisioned VM's. After migrating or deleting VM's there will be a lot of "small" free blocks all over the VMFS. In this case, when trying to create a thick disk on this VMFS, will this thick disk be created using the distributed free blocks and therefore be fragmented (assuming none of the free spaces is big enough for the whole VMDK) or does ESX refuse to create the disk?
As I mentioned above, I'd be interested in your opinions.