What is "Performance Metadata" shown in the "volume show-space" output?

Applies to

• Clustered Data ONTAP 8.3
• ONTAP 9

Answer

What does Performance Metadata stand for?

Performance Metadata is a collection of the blocks that are consumed by the deleted/overwritten data, these blocks haven't been accounted for in the free space as their corresponding bitmaps are yet updated. It's introduced by a new feature called  Volume Blog.

What is "Volume Blog"?

• "Volume Blog" is a new featured introduced from Data ONTAP 8.3, it's used to accumulate the bitmap updates to lessen the performance impact caused by random bitmap updates. 
• With this new feature, the second half of the block freeing procedure might be postponed, and all the postponed bitmap updates would be done in a batcheing manner at a later point.

What is the procedure to free blocks at the Volume layer?

Two steps have to be carried out to clear up a file and reclaim the space. For example, if a file is to be deleted:

• WAFL has to traverse the buftree of this file, and tear down all of indirect blocks.
• WAFL has to update the bitmaps to mark those deleted blocks(both indirect blocks and the user data blocks) as free blocks.
  • Note: The file is already gone after the first step is done, but the space will only be reclaimed after the second step is complete. 

What is the difference in space reporting that "blog" brings to us? 

• The space occupied by the deleted files might not be able to return to the free space immediately.

Note: The free space could be checked in the available column of the df vol_name output

• There is a pre-defined threshold, which is 1.2% of the volume size. Once the Blog file size reaches this 1.2% threshold, WAFL will start draining it right away and return it to the free space in a batch.
  • In most cases, the Blog file size, or the space locked by Volume Blog would be lower than this 1.2% threshold. 
  • But if the rate of new deletions inserted into the Volume Blog is faster than the rate it gets drained from the Blog, then Blog file would continue to grow until it reaches the Blog Upper Cap.
     
• There is an Upper Cap for the space locked in this Performance Metadata, once the Upper Cap is breached, all the subsequent deletions would be carried out in a synchronous way(In CPs) without being attached to Blog file. 
  • Upper Cap for Overwrites : 3%
  • Upper Cap for Hole Punch(SCSI UNMAP): 2%
  • Upper Cap for Zombie: 5%

Note: All the above examples are against the file deletion, but Volume Blog works in a similar way for overwrites and hole punching. 

Additional Information

Add your text here.