During a consistency point, the data is written from memory to disk. This action requires calculating the RAID (XOR processing). The XOR processing is CPU intensive and is carefully calibrated by ONTAP. This is called CP pacing.
CP pacing controls the XOR batch processing size, which is the main CPU factor with regards to RAID processing for CP.
If the CP pacing increases the batch size then XOR threads consume more CPU and process data faster. Thus a tetris is handed off to storage layer quicker.
CP pacing also looks at the outstanding blocks required to be written for the current CP versus the resource fill for the next CP to adjust the batch size. For ex: When blocks written is less than the next CP fill , the batch size is increased.
The CP pacing batch size varies from 1 to 18.
Thus, cp_batch_override 18 means run XOR as fast as possible, which should decrease the P2_FLUSH time, assuming there is no actual storage issue.
CP pacing controls the XOR batch processing size, which is the main CPU factor with regards to RAID processing for CP.
If the CP pacing increases the batch size then XOR threads consume more CPU and process data faster. Thus a tetris is handed off to storage layer quicker.
CP pacing also looks at the outstanding blocks required to be written for the current CP versus the resource fill for the next CP to adjust the batch size. For ex: When blocks written is less than the next CP fill , the batch size is increased.
The CP pacing batch size varies from 1 to 18.
Thus, cp_batch_override 18 means run XOR as fast as possible, which should decrease the P2_FLUSH time, assuming there is no actual storage issue.
No comments:
Post a Comment