Thanks! I see you have found this post - https://community.netapp.com/t5/ONTAP-Discussions/Automatic-node-referrals-CIFS-and-support-for-Hyper-V/td-p/147402 - I have not replied to that one as it is so old, but I think their problem was that they only have data volumes/aggregates on one node.
The current link to our best practices document for SMB for HyperV is at https://docs.netapp.com/us-en/ontap/pdfs/sidebar/SMB_configuration_for_Microsoft_Hyper_V_and_SQL_Server.pdf
As your system is seeing indirect access, I assume that the volumes and aggregates are spread. With at least 10Gb throughput, on 9.9.1 indirect access should not be a major concern.
I can't say for sure why node referrals is required to be disabled for Hyper-V, but I suspect it is a requirement for continuous availability, as this would otherwise remove indirect access. I believe therefore that the indirect access is expected in this use case.
You can run this command to view latency sources in your environment - I would be interested to see the output:
cluster1::> qos statistics volume latency show -iterations 100
Workload ID Latency Network Cluster Data Disk QoS NVRAM Cloud
--------------- ------ -------- -------- -------- -------- -------- -------- ---------- ----------
-total- 110.35ms 0ms 110.02ms 327.00us 0ms 0ms 0ms 0ms
vs1vol0 111 167.82ms 0ms 167.22ms 603.00us 0ms 0ms 0ms 0ms
vol1 1234 117.76ms 0ms 117.56ms 191.00us 0ms 0ms 0ms 0ms
