QoS Limit Max

The maximum latency ceiling defined by the QoS policy. When the latency line approaches or touches this line, ONTAP is throttling the workload, and the QoS limit—rather than a physical resource—is the dominant source of latency.

QoS Limit Min

The guaranteed minimum latency floor defined by the QoS policy. This line indicates the response time floor enforced for the workload. When other workloads use QoS minimums to guarantee their throughput, ONTAP can throttle this workload, which then shows higher latency.

Read latency

Average time to complete a client read request on the volume, from when the request is received until data is returned. Read requests must traverse the full I/O path from the network protocol layer through the data processing stack to the aggregate where the data resides.

Aggregate or disk contention; cache misses that promote reads to physical disk; cold data retrieved from a cloud capacity tier via FabricPool; cross-node reads when data resides on a different node than the one handling the request.

Write latency

Average time to acknowledge a client write request on the volume. ONTAP acknowledges a write once it is committed to the NVRAM write log; the data is later flushed to the aggregate.

NIC saturation or high round-trip latency between the client and the storage node; synchronous SnapMirror waiting for the destination cluster to confirm receipt before the write can be acknowledged; NVRAM pressure under heavy write loads; CPU overhead from inline deduplication, compression, or compaction running in the write I/O path.

Other latency

Average time to complete non-read, non-write operations on the volume, including file opens, attribute lookups, directory traversals, file creates, and locks. Elevated other latency can affect application responsiveness even when read and write latency appear normal.

CPU pressure from inline efficiency operations competing with metadata processing; high namespace activity from workloads that generate large numbers of file creates, deletes, or directory scans; QoS throttling that constrains total IOPS, leaving fewer IOPS available for metadata operations; volume activation overhead when many volumes are concurrently active.