Bursty write workload could affect performance

Applies to

• ONTAP 9.x
• ONTAP 8.x

Issue

WAFL(ONTAP file system) processes data well when the incoming data flow is steady. However, bursty write workload could overwhelm the processing layer and lead to B/b(Back to Back) CPs and high write latency. 

• How to identify a bursty workload?
  • Bursty workload can be checked using the sysstat -x command from the node shell, which provides output per one-second intervals.
  • Significantly fluctuated incoming data between the one-second iterations is indicative of bursty workload.
  • Check 'Net In' column in case of NFS, CIFS.
  • Check 'Net In' or 'iSCSI In' column in case of iSCSI.
  • Check 'FCP In' column in case of FCP.
  • Much higher CPU util could also be observed along with the write workload bursts.

Example:

• How to identify the high latency caused by bursty write workload?

High latency from Data column in the qos statistics workload latency show or qos statistics volume latency show outputs.

Example:

cluster01::> qos statistics workload latency show -iterations 1 -rows 3
Workload            ID    Latency    Network    Cluster       Data       Disk    QoS Max    QoS Min      NVRAM      Cloud  
--------------- ------ ---------- ---------- ---------- ---------- ---------- ---------- ---------- ---------- ---------- 
vol1             21879    287.63ms     118us        0ms   286.00ms   589.00us        0ms        0ms     1.00us        0ms