What are IOPs?

All storage systems

IOPs stands for Input/output Operations Per Second, and are each SCSI command or NAS operation
Commonly used to measure the performance of storage
For read or write operations, please keep in mind they can be measured in units of bytes
- Sometimes this is called operation or block size
- A 1 MB file for example can be read in 256 4kB blocks, or 16 64kB blocks
  - In this example, 4kB would be 256 read IOPs, or 16 64kB read IOPs
Other IOPS are anything not read/write (metadata generally) and each request on the wire is accounted as one IOP in ONTAP

Random IOPs will take more CPU and may have higher latency than sequential
- Random IOPs blocks are read not in any order off of disk, like block 3, 12, 66, 1, etc.
- Sequential IOPs blocks are read in order off of disk, like block 1, 2, 3, 4, etc.
Smaller operation sizes generally require more computation
- In the example above of a 1 MB file, ONTAP has to go to disk 256 times and process 256 operations at 4kB vs only 16 at 64kB
ONTAP will handle over 64kB fine, but due to the normalization of operation sizes, latency may appear slightly higher
- This is better in newer versions of ONTAP 9 compared to Data ONTAP 8
- A 1 MB read size will have 16 WAFL requests of 64kB each in parallel

No
- Disk IOPS vary for various reasons and are decoupled from backend IOPS
- For more information see Why does the sum of all volume IOPs in an aggregate not match the aggregate IOPs?

ONTAP does not track IOPS for internal processes but uses different methods of accounting them.
IOPS are generated from user requests for all served data protocols (CIFS/FCP/iSCSI/NFS/NVMe*/S3).
- The only exception is if you use set advanced; statistics start -object volume -counter repl_*_ops which will show SnapMirror IOPS.
- The commands for qos statistics and the workload objects for statistics, as well as other commands are all user IOPS.
If a workload burst happens, then it must be investigated from the client end.