You have an ONTAP 2-node cluster.

The QoS latency output indicates the problem is associated with the HA interconnect/NVRAM mirroring path rather than ordinary disk latency. In ONTAP HA pairs, writes are protected by logging them locally and mirroring the write log to the partner. If the HA interconnect path is impaired, write acknowledgement latency can rise even when front-end network latency and disk latency do not dominate the workload. Increasing SMB read sizes would not address the observed write-path protection delay. Replacing NVRAM is too aggressive without evidence of a failed NVRAM device, and a disk bottleneck would show disk latency as the dominant center. The support engineer should interpret the QoS columns as latency centers and then correlate them with HA architecture. When NVRAM-related latency is elevated in an HA pair, HA interconnect health, cabling, ports, and partner connectivity must be reviewed before assuming a media or protocol issue. References/topics: ONTAP QoS latency centers, HA interconnect, NVRAM mirroring, SMB workload troubleshooting, and write-path latency analysis.