MicrosoftThanks for this great article.
We have a Windows Server 2019 Hyper-V 4 node, storage spaces direct cluster with 2 dual port 25Gbps NICs.
Two of the 25G NICs are in a SET team and have 3 vNICs (vCluster, vLiveMigration and vBackup)
The other 2 x 25G NICs are not teamed and have IP addresses configured in different subnets. Our idea here was to dedicate these 2 physical NICs to be used for storage traffic only. This was because we took the online recommendations that Microsoft say to use iWARP to save complexity of setting up QoS. As we are not doing QoS we though dedicate 2 NICs to storage only. We didn't think the overhead of creating a SET Team would have any significant advantages over just using SMB Direct and multichannel on the physical NICS so they were not teamed.
BUT, I cannot see these SMB NICs being used even when large file copy traffic to the test VM is generated. I do see extra traffic on the SET team interfaces but this is not what we intended. Obviously we are doing something wrong. See the cluster network metrics below in case helps this explain anything.
Points to Note:
1. I was using perf mon RDMA activity for the monitoring of storage traffic.
2. Also in perfmon when selecting 'SMB Direct Connection' it doesn't give the choice for the SMB interfaces, it only shows the virtualised interfaces created on the SET team.
3. The VM is using a 4 x 1Gbps SET team for its traffic.
I feel there is an important concept missing in understanding how Hyper-V 2019 selects its storage network. Appreciate any comments. Thanks.
