What are the performance considerations of deduplication in ONTAP systems?

Applies to

• ONTAP 9
• FAS
• AFF
• Cloud Volumes ONTAP

Answer

• Avoid deduplication of datasets with a high rate of change. I.e data that is overwritten every few days
• Avoid running deduplication operations on vols where efficiency gains are less than ~10%.

Example: The datastore volume has an efficiency gain well above 10%, in this case 55% using the df -s command

::> df -s datastor​e​​​​
Filesystem                used      saved       %saved  Vserver
/vol/datastore/      289150724  356485168          55%  svmtest​​​

• Inline efficiencies also have a latency penalty for each read or write to hard drive systems (FAS)
  • FAS systems are designed for inline efficiencies to be off by default due to a severe cost of latency from disk.
  • AFF or Capacity Flash are built around this as required
• Ensure monitoring of performance is enabled so performance when deduplication runs can be monitored on CPU/disk latencies compared to when it is idle.
  1. ActiveIQ Unified Manager is a free product and can be deployed in the same network as your storage.
  2. The ActiveIQ portal does have performance graphs, but are very high level.

Additional Information

TR-3966 has more information about deduplication in FAS and AFF systems