Clustered ONTAP 8.2/8.3
QOS / Adaptive QOS
- Most commonly its the perception that QOS is limiting workload too aggressively.
- This is partly due to aggregating performance counters reported over a defined time-period, typically from 5 minutes to as low as 3 seconds with the
qos statisticscommand. Which in the case of bursty workloads can be far too coarse.
- For example, if a QOS limit applied max throughput of 100 IOPS. However, if 120 ops are received within 1 second and is effectively idle for remainder of the time sample.
- In such cases, even at 3s granularity, would report low IOPs (significantly less than the 100 IOPs limit), in conjunction with QOS latency. As such, can appear that QOS latency is being induced before appropriate limits are hit.
- This article outlines how to validate if workload is bursting too high for the applied qos ratebucket limits.