Hello,
The intermittent write drops you’re seeing are a classic symptom of SMB Direct traffic competing with regular host traffic on converged NICs. To stabilize throughput, you need to enforce QoS policies that explicitly prioritize SMB Direct (RDMA) traffic. Windows Server supports this through Data Center Bridging (DCB) and QoS policies bound to SMB ports.
The recommended approach is to configure QoS at the host level using PowerShell. First, identify the SMB Direct ports in use (445/TCP for SMB, plus RDMA traffic over iWARP/RoCE). Then create a QoS policy that matches this traffic and assigns it a higher priority. For example:
New-NetQosPolicy -Name "SMBDirect" -SMB -PriorityValue8021Action 3
This command creates a QoS policy specifically for SMB traffic and maps it to IEEE 802.1p priority 3. You can adjust the priority value depending on your network design, but typically SMB Direct should be placed above default OS traffic. If you’re using RDMA over RoCE, you must also configure DCB with Priority Flow Control (PFC) to ensure lossless transport. That means enabling PFC on the NIC and switch ports for the same priority you assigned to SMB Direct.
Verify the configuration with:
Get-NetQosPolicy
Get-NetQosFlowControl
On the switch side, ensure that the QoS trust boundary is set to honor 802.1p markings from the host. Without this, the prioritization won’t propagate through the fabric. Also confirm that other traffic classes (like cluster heartbeat or management) are assigned lower priorities so they don’t interfere with SMB Direct.
In short, the fix is to create a host-level QoS policy for SMB Direct, align it with DCB/PFC settings, and validate that the network fabric respects those markings. This ensures your Scale-Out File Server shares deliver consistent throughput even under heavy load.
I hope you've found something useful here. If it helps you get more insight into the issue, it's appreciated to accept the answer. Should you have more questions, feel free to leave a message. Have a nice day!
HP.