Why is Volume write throughput much higher than Workload write throughput when SCSI WRITE SAME is present
Applies to
- ONTAP 9
- AIQ PAS
- SCSI WRITE SAME
Answer
This is the expected behavior.
- SCSI WRITE SAME is a SCSI operation that tells the storage to write a certain pattern a number of times to consecutive blocks.
- In most cases, SCSI WRITE SAME is used to write all zeros so the client can initialize the data on block devices.
- Instead of issuing a large amount of zeros, the client only has to issue a few hundred or thousand small WRITE SAME operations, that's why this payload will NOT be charged at the Workload or LUN layer.
- ONTAP will convert these WRITE SAME operations to a lot more WAFL WRITE messages, that's why the payload/throughput is observed with Volume layer.
Additional Information
SCSI Copy Offload works in a similar way and might also lead to a discrepancy in throughput between the Workload layer and the Volume layer.
- ONTAP has per-volume statistics to track down the Copy Offload opeations.
- Copy Offload will cause a discrepancy for both the Write Ops/Throughput and the Read Ops/Throughput, depending on it's the Copy Offload source or destination.