We have ONTAP 8.1.2P4 and ONTAP 8.1.4P2 in our environment. We have a single thin provisioned LUN contained in a thin provisioned volume. The LUN is used by ESXi 5.1 as a VMFS5 formatted datastore. The LUN type is vmware.
We just noticed that the volume IOP statistics show materially higher numbers than the LUN IOP statistics. For example, the LUN may show 150 write IOPS and the volume shows 225 write IOPS. My expectation is that the volume IOPS would be the same as the LUN IOPS given the LUN is the only object in the volume.
Is it normal to see higher volume IOP numbers than LUN IOP numbers in this situation? We suspect the additional IOPS captured at the volume may be a result of thin provisioned space being written to within the LUN and/or volume. The VMFS5 datastore should be aligned properly as the LUN type is vmware however the virtual machines on the datastore appear to perform partial-writes. Could the partial-writes from the virtual machines be to blame for the discrepancy between volume and lun IOPS statistics?
Just curious if anybody else has seen similar behavior or if this is expected.
Update: Another thought is that the difference is related to the disk provisioning being Thick Provisioned Lazy Zeroed. " A lazy zeroed disk is not zeroed up front, which makes the provisioning very fast. However, there is an added latency on first write because each block is zeroed out before it is written to for the first time." So perhaps the added write IO we are seeing is related to the zero'ing process?