Thin Provisioning is a vast and crucial topic. I do not think we can cover everything about Thin Provisioning over here. I am trying to put few points about it.
Thin Provisioning is achieved by the combination of THIN & DATA devices.
In Thin Provisioning, storage (Thin devices) is created and allocated from a shared Thin pool (combination of Data devices). Thin device should be bound to Thin pool before it is used by Host. Host treat the Thin Devices(TDEV) as any other device. Until an application on Host writes to the Thin Device no physical storage is consumed. If an application on the host writes 1GB of data to the Thin Device, 1GB of actual storage is used from the Data Devices in the Thin pool. No actual data is saved on Thin device. This type of allocation is meant for predictable/controllable applications.
When a Thin device writes to a Thin Pool the data is striped across the Data devices. The more Data devices in the pool, the wider the striping. This will improve performance. Have as many as Data devices (smaller size) in the Thin pool. This will help to stripe the I/O across many spindles which will eventually increase the performance.
We have great Guns in this forum. Hope they will shed some light on this.
GEMC
36 Posts
0
June 6th, 2011 11:00
Thin Provisioning is a vast and crucial topic. I do not think we can cover everything about Thin Provisioning over here. I am trying to put few points about it.
Thin Provisioning is achieved by the combination of THIN & DATA devices.
In Thin Provisioning, storage (Thin devices) is created and allocated from a shared Thin pool (combination of Data devices). Thin device should be bound to Thin pool before it is used by Host. Host treat the Thin Devices(TDEV) as any other device. Until an application on Host writes to the Thin Device no physical storage is consumed. If an application on the host writes 1GB of data to the Thin Device, 1GB of actual storage is used from the Data Devices in the Thin pool. No actual data is saved on Thin device. This type of allocation is meant for predictable/controllable applications.
When a Thin device writes to a Thin Pool the data is striped across the Data devices. The more Data devices in the pool, the wider the striping. This will improve performance. Have as many as Data devices (smaller size) in the Thin pool. This will help to stripe the I/O across many spindles which will eventually increase the performance.
We have great Guns in this forum. Hope they will shed some light on this.