Task 02: Configure Dell PowerSwitch
DOCUMENT CATEGORY: Runbook
SCOPE: Top-of-Rack switch configuration
PURPOSE: Configure Dell switches for RDMA, VLANs, and high availability
MASTER REFERENCE: Dell PowerSwitch OS10 Documentation
Status: Active Estimated Time: 90 minutes Last Updated: 2026-11-15
Overview
Configure Dell PowerSwitch TOR switches with Data Center Bridging (DCB) for RDMA (RoCE), Virtual Link Trunking (VLT) for redundancy, and VLANs for network segmentation.
The toolkit keeps no finished switch configuration. config/network-devices/switches/dell/tor.os10.template is the configuration of a Dell OS10 top-of-rack switch with every site value as a placeholder named after the infrastructure.yml key it comes from, and New-SwitchConfiguration.ps1 fills it once per switch. The earlier hand-edited files tor1-azlocal-config.txt and tor2-azlocal-config.txt are gone.
Prerequisites
| Requirement | Description |
|---|---|
| OpenGear Access | Step 1 completed - console access available |
| Switch Console | Access via OpenGear serial ports or SSH |
| Network Design | The switches, their ports and the VLANs in infrastructure.yml (keys below) |
| Storage VLANs | The storage networks of networking.storage.networks[] available (non-routable) |
| Firmware | Dell-validated firmware version installed |
Values from infrastructure.yml
New-SwitchConfiguration.ps1 renders each entry of network_devices[] whose type is switch and whose vendor has a template (Dell: tor.os10.template). It reads these keys from infrastructure.yml (schema 5). Each key, with its type, default and example, is in the variables reference.
| Part of the configuration | infrastructure.yml keys |
|---|---|
| Hostname, management address | network_devices[].hostname, network_devices[].management.ip_address |
| Management prefix and route, SSH source | networking.oob.subnet, networking.oob.gateway |
| Time zone (UTC offset of the Windows time zone ID) | site.time_zone |
| VLANs | networking.management.vlan_id, network_devices[].ports.native_vlan_id, networking.oob.vlan_id, networking.logical_networks[].vlan_id, networking.storage.networks[] (only with scenario.storage_connectivity: switched) |
| QoS (PFC and ETS) | The intent of networking.intents[] with storage traffic: overrides.qos.cluster_priority, smb_priority, smb_bandwidth_percent (registry defaults 7, 3 and 50: Network ATC); the cluster reservation is 1 percent when every storage adapter is faster than 10 Gbps, else 2 (hardware.network_cards[].ports[].speed_gbps) |
| Ports, MTU | network_devices[].ports.uplinks, .node_ports, .peer_link, network_devices[].mtu |
| VLT pair | network_devices[].ha.domain_id, .peer_address, .priority, .system_mac |
| DNS and NTP | networking.management.dns_servers, networking.ntp_servers |
| Login | network_devices[].credentials.username; the password is a secret reference that the script names and never reads |
The keys added in registry 5.0.62 (infrastructure schema 5.6.28, all optional):
| Key | What it is | Example |
|---|---|---|
network_devices[].mtu | Maximum frame size of the ports that face the machines, the uplinks and the link between the two switches of a pair, in bytes | 9216 |
network_devices[].ports.uplinks | Ports to the network above the rack (trunks with the management, out-of-band and workload VLANs) | [ethernet1/1/1] |
network_devices[].ports.node_ports | Ports cabled to the machines' network adapters (trunks with the VLANs of the intents; with RDMA storage on the switch they get the PFC and ETS policies) | [ethernet1/1/2, ethernet1/1/3] |
network_devices[].ports.peer_link | Ports of the link between the two switches of a pair (OS10: VLTi, the vlt-domain discovery interfaces) | [ethernet1/1/13, ethernet1/1/14] |
network_devices[].ports.native_vlan_id | The VLAN the switch puts the untagged frames of the node ports in (OS10: switchport access vlan on a trunk port) | 2201 |
network_devices[].ha.domain_id | The ID of the high availability domain the two switches of a pair share (OS10: vlt-domain) | 1 |
network_devices[].ha.peer_address | The management address of the other switch of the pair, its backup heartbeat (OS10: backup destination) | 198.51.100.7 |
network_devices[].ha.priority | Election priority of this switch in the pair; the lower value becomes the primary (OS10: primary-priority) | 100 |
network_devices[].ha.system_mac | The MAC address the pair presents as one system (OS10: vlt-mac) | 00-00-5E-00-53-01 |
A key the file lacks falls back to its registry default; a key with no default that the template needs stops the run with its name. Port names differ by switch model, so they are keys, not part of the template.
Generate the switch configurations
# Render and check every switch, write nothing.
./scripts/deploy/03-onprem-readiness/phase-03-network-infrastructure/task-02-dell-powerswitch-configuration/powershell/New-SwitchConfiguration.ps1 `
-InfrastructureFile ./infrastructure.yml -OutputDirectory ./switch-config -WhatIf
# Write <hostname>.cfg for each switch (or only some with -Hostname).
./scripts/deploy/03-onprem-readiness/phase-03-network-infrastructure/task-02-dell-powerswitch-configuration/powershell/New-SwitchConfiguration.ps1 `
-InfrastructureFile ./infrastructure.yml -OutputDirectory ./switch-config
| Parameter | What it is |
|---|---|
-InfrastructureFile | The master infrastructure.yml (schema 5). Only the keys the template names are read |
-OutputDirectory | Where the configurations go, one <hostname>.cfg per switch. Keep it out of the repository: the files hold the site's addresses |
-Hostname | Optional. Only the switches with these host names |
-TemplateRoot | Optional. The folder with one folder per vendor (default: config/network-devices/switches of the toolkit) |
-WhatIf | Renders and checks everything and writes nothing |
The script never connects to a switch, a node or Azure, and no password is read or written: network_devices[].credentials.password must be a secret reference (keyvault://, env://, file:// or command://), and the output only tells you to create the account with the password from the secret store. A switch whose vendor has no template is reported and skipped.
It stops instead of rendering a wrong switch: an SMB bandwidth that leaves nothing for the other traffic, equal SMB and cluster priorities, or a template placeholder that names a password, secret or token.
| Exit code | Meaning |
|---|---|
0 | Every switch was written (or rendered with -WhatIf), or the file lists none |
1 | A file could not be written, or a rendered configuration still holds a placeholder |
2 | A value in the file is missing or not usable; the message names the key |
The network team reviews each file and applies it on its switch (console or SSH: configure terminal, paste, show running-configuration, write memory).
Configuration Areas Overview
What the template configures, and where each part takes its values:
| Configuration Area | Purpose | Key Settings |
|---|---|---|
| System Settings | Basic switch identity and management | Hostname, management IP on the out-of-band network, time zone, DNS, NTP, SSH |
| QoS/DCB for RDMA | Enable lossless storage traffic | DCBX, PFC on the SMB priority, ETS bandwidth of Network ATC |
| VLT (Virtual Link Trunking) | Switch redundancy and high availability | VLT domain, discovery interfaces, backup destination, primary priority, VLT MAC |
| VLANs | Network segmentation | Management, out-of-band, workload and (switched storage) storage VLANs |
| Trunk Ports | Node connectivity | MTU, QoS policies, allowed VLANs, untagged VLAN |
| Uplinks | Core/distribution connectivity | Trunks with the management, out-of-band and workload VLANs |
Values that stay in the template are fixed by Dell or Microsoft and say so in a comment: the SSH ciphers, LLDP and DCBX, the console and session hardening, RSTP with bridge priority 4096, the OS10 syntax. Syslog and SNMP forwarding are parked: the template leaves them out, and they have no keys.
Section A: QoS/DCB Configuration for RoCE
Azure Local requires Data Center Bridging (DCB) with Priority Flow Control (PFC) and Enhanced Transmission Selection (ETS) for reliable RDMA over Converged Ethernet (RoCE). The template writes this section only when scenario.storage_connectivity is switched and an intent carries storage traffic.
Network ATC configures the hosts with cluster heartbeat priority 7 (1 percent when the adapters are faster than 10 Gbps, 2 percent otherwise), SMB Direct priority 3 with 50 percent, and priority 0 for the rest, and Microsoft says the physical network must match (Network ATC default values). The template takes the same numbers from the file, so a change in networking.intents[].overrides.qos changes the hosts and the switches. Dell's older guides use priority 5 for the cluster class; the toolkit follows Network ATC.
Key Configuration Elements
| Element | Purpose | Configuration |
|---|---|---|
| DCBX | DCB exchange protocol | Enable globally |
| Trust Map | Map 802.1p priorities to queues | SMB priority (default 3), cluster priority (default 7), the rest to queue 0 |
| PFC Policy | Lossless traffic for SMB | Pause on the SMB priority |
| ETS Policy | Bandwidth allocation | SMB smb_bandwidth_percent (default 50), cluster 1 or 2 percent, the rest to the default class |
| Jumbo Frames | Large frame support for storage | network_devices[].mtu on the node ports, uplinks and VLTi |
QoS as rendered with the defaults (OS10)
With the registry defaults and storage adapters faster than 10 Gbps, the QoS part of a generated file reads:
class-map type queuing QDefault
match queue 0
class-map type queuing QSmb
match queue 3
class-map type queuing QCluster
match queue 7
class-map type network-qos SmbStorage
match qos-group 3
class-map type network-qos ClusterHeartbeat
match qos-group 7
trust dot1p-map trust_map
qos-group 0 dot1p 0-2,4-6
qos-group 3 dot1p 3
qos-group 7 dot1p 7
qos-map traffic-class queue-map
queue 0 qos-group 0-2,4-6
queue 3 qos-group 3
queue 7 qos-group 7
policy-map type queuing ets-policy
class QDefault
bandwidth percent 49
class QSmb
bandwidth percent 50
class QCluster
bandwidth percent 1
policy-map type network-qos pfc-policy
class SmbStorage
pause
pfc-cos 3
system qos
trust-map dot1p trust_map
Section B: VLT Configuration for High Availability
VLT (Virtual Link Trunking) provides switch redundancy with active-active uplinks. The template writes the VLTi ports and the VLT domain only when network_devices[].ha.domain_id is set.
VLT Components
| Component | Description | Key |
|---|---|---|
| VLT Domain | Logical grouping of two VLT peer switches | network_devices[].ha.domain_id |
| VLT Peer Link (VLTi) | Inter-switch link for control and data sync | network_devices[].ports.peer_link |
| Backup Destination | Out-of-band heartbeat via the management interface of the other switch | network_devices[].ha.peer_address |
| Primary Priority | The lower value becomes the primary | network_devices[].ha.priority |
| VLT MAC | Virtual MAC address for the VLT domain | network_devices[].ha.system_mac |
VLT domain as rendered (OS10)
With the example values of the variables reference (documentation addresses), one switch of the pair:
interface ethernet1/1/13
description VLTi
no shutdown
no switchport
mtu 9216
flowcontrol receive off
flowcontrol transmit off
priority-flow-control mode on
service-policy input type network-qos pfc-policy
service-policy output type queuing ets-policy
ets mode on
qos-map traffic-class queue-map
exit
vlt-domain 1
backup destination 198.51.100.7
discovery-interface ethernet1/1/13
discovery-interface ethernet1/1/14
peer-routing
primary-priority 100
vlt-mac 00:00:5e:00:53:01
exit
The other switch has its own network_devices[] entry: the same domain_id and system_mac, the first switch's management address as peer_address, and a higher priority.
Management Interface
interface mgmt1/1/1
no shutdown
no ip address dhcp
ip address <network_devices[].management.ip_address>/<prefix length of networking.oob.subnet>
no ipv6 enable
exit
management route 0.0.0.0/0 <networking.oob.gateway>
SSH is allowed only from networking.oob.subnet (access list SSH-Allowed on the vty lines).
Section C: VLAN Configuration
VLAN Types
| VLAN Type | Purpose | Layer 3 Routing | Key |
|---|---|---|---|
| Management VLAN | Node management | Yes (routable) | networking.management.vlan_id |
| Untagged VLAN of the node ports | Untagged frames of the machines' adapters | network_devices[].ports.native_vlan_id | |
| Storage VLANs | SMB storage traffic between nodes | No (east/west only) | networking.storage.networks[].vlan_id |
| Workload VLANs | VM workload networks | Yes (typically routable) | networking.logical_networks[].vlan_id |
| Out-of-band VLAN | iDRAC/BMC and switch management | Yes (routable, isolated) | networking.oob.vlan_id |
The storage VLANs are non-routable and used exclusively for east/west traffic between Azure Local cluster nodes. These VLANs:
- Are configured on ToR switch trunk ports connecting to cluster nodes (the template adds them only with
scenario.storage_connectivity: switched) - Are NOT configured on firewalls or any Layer 3 interfaces
- Are managed automatically by Microsoft Network ATC during cluster deployment
- MUST be verified as available in the customer's environment during discovery
The template creates one interface vlan per unique VLAN ID of these keys, and one per storage network with its name as the description.
Section D: Trunk Port Configuration for Cluster Nodes
Each port of network_devices[].ports.node_ports becomes a trunk:
interface ethernet1/1/2
description AzureLocalNode
no shutdown
switchport mode trunk
switchport access vlan <network_devices[].ports.native_vlan_id>
switchport trunk allowed vlan <management and workload VLANs, without the untagged VLAN>
switchport trunk allowed vlan <storage VLANs (switched storage only)>
mtu <network_devices[].mtu>
flowcontrol receive off
flowcontrol transmit off
priority-flow-control mode on
service-policy input type network-qos pfc-policy
service-policy output type queuing ets-policy
ets mode on
qos-map traffic-class queue-map
spanning-tree bpduguard enable
spanning-tree port type edge
exit
The PFC and ETS lines are written only with switched storage and a storage intent. A VLAN is either the untagged VLAN of a port or one of its tagged VLANs, not both.
Section E: Uplink Configuration
Each port of network_devices[].ports.uplinks becomes a trunk with the management, untagged, out-of-band and workload VLANs:
interface ethernet1/1/1
description Uplink
no shutdown
switchport mode trunk
switchport trunk allowed vlan <the VLAN list>
mtu <network_devices[].mtu>
flowcontrol receive on
flowcontrol transmit off
exit
Section F: Services
The template enables LLDP, DCBX and SSH (with the ciphers and MAC the toolkit allows), sets the time zone from site.time_zone, the DNS servers from networking.management.dns_servers, the NTP servers from networking.ntp_servers, RSTP with bridge priority 4096, and the local audit log, and ends with write memory. The login account network_devices[].credentials.username is written as a comment: create it on the switch with the password from the secret store (username <name> password <password> role sysadmin).
Post-Configuration Verification
1. Verify VLT Status
show vlt 1
Expected output:
- Both switches show VLT domain as "Up"
- Peer link status: "Up"
- Backup link status: "Up"
2. Verify QoS Configuration
show running-configuration class-map
show running-configuration policy-map
show qos interface ethernet1/1/1
- Confirm PFC enabled on the SMB priority (default 3)
- Confirm ETS bandwidth allocation matches Network ATC (SMB
smb_bandwidth_percent, cluster 1 or 2 percent) - Confirm trust map applied
3. Verify VLANs
show vlan
- All required VLANs present and active (the VLAN IDs of the keys in Section C)
4. Verify Port Status
show interface status
- All node-facing ports up with correct speed
Validation Checklist
- VLT domain operational (both switches)
- VLT peer link and backup link up
- Configuration generated with
New-SwitchConfiguration.ps1frominfrastructure.ymland reviewed - QoS/DCB configured (DCBX, PFC on the SMB priority, ETS of Network ATC)
- Trust map applied system-wide
- All VLANs active (management, out-of-band, workload, storage)
- Trunk ports configured with jumbo frames (
network_devices[].mtu) - All node-facing ports up with correct speed
- Uplinks operational
- LLDP enabled
- SSH enabled
- Configuration saved to startup
Troubleshooting
| Issue | Possible Cause | Resolution |
|---|---|---|
| VLT peer link down | Incorrect port-channel member assignment or cable fault | Verify VLT configuration, check cables between switches, confirm port-channel membership |
| PFC not negotiating | DCBX mode mismatch between switch and NIC | Set DCBX to IEEE on both ends, verify NIC firmware supports RoCEv2 |
| Storage VLAN unreachable | VLAN not added to trunk port or VLAN not created | Verify VLAN exists (show vlan), confirm trunk port has VLAN tagged |
| Jumbo frames not working | MTU mismatch along path | Confirm network_devices[].mtu (for example 9216) on all switch ports and the host NICs in the storage path |
New-SwitchConfiguration.ps1 ends with exit code 2 | A key the template needs has no value and no default, or the QoS of the storage intent is not usable | Set the key the message names in infrastructure.yml and run it again |
| PFC priority differs between host and switch | The switch was configured from an older guide (cluster priority 5) | Generate the configuration again: it follows networking.intents[].overrides.qos and Network ATC |
| SSH access denied | SSH service not enabled or management ACL blocking | Run show ip ssh, verify SSH enabled, check management ACL entries |
Next Steps
Proceed to Task 3 - Verify Firewall Endpoints to verify firewall rules for required Azure and Dell endpoints.
- Manual
- Orchestrated Script
When to use: Use this option for manual step-by-step execution.
See procedure steps above for manual execution guidance.
PowerShell
When to use: Run the task from a management workstation or a pipeline. The scripts read
infrastructure.ymland carry no values of their own.
Scripts
| Script | What it does |
|---|---|
scripts/deploy/03-onprem-readiness/phase-03-network-infrastructure/task-02-dell-powerswitch-configuration/powershell/New-SwitchConfiguration.ps1 | Writes the configuration of every top-of-rack switch of network_devices[] from the vendor's template (config/network-devices/switches/dell/tor.os10.template), one <hostname>.cfg each. Sends nothing to a switch. Supports -WhatIf. |
scripts/deploy/03-onprem-readiness/phase-03-network-infrastructure/task-02-dell-powerswitch-configuration/powershell/Set-PowerSwitchEndpoint.ps1 | Checks that every switch of network_devices in infrastructure.yml can be reached and that its login can be read. Changes nothing on a switch. |
Requires: PowerShell 7.
Run
./New-SwitchConfiguration.ps1 -InfrastructureFile ./infrastructure.yml -OutputDirectory ./switch-config -WhatIf
./New-SwitchConfiguration.ps1 -InfrastructureFile ./infrastructure.yml -OutputDirectory ./switch-config
./Set-PowerSwitchEndpoint.ps1 -InfrastructureFile ./infrastructure.yml
Scripts for this task are located in the azurelocal-toolkit repository under scripts/deploy/ in the appropriate task folder.
Alternatives
The procedures in this task use the methods shown in the tabs above. The toolkit is PowerShell 7 only and has no Bash scripts. A few tasks that act on Azure also have an Azure CLI script written in PowerShell. The script status page of the toolkit lists which task has which script.
Navigation
| ← Task 01: OpenGear Configuration | ↑ Part 3: On-Premises Readiness | Task 03: Firewall Endpoint Verification → |
Version Control
| Version | Date | Author | Changes |
|---|---|---|---|
| 1.0 | 2026-01-31 | Azure Local Cloud CI/CD Engineering | Initial document |
| 1.1 | 2026-03-03 | Azure Local Cloud CI/CD Engineering | Standardized runbook format |
| 1.2 | 2026-11-15 | Azure Local Cloud CI/CD Engineering | The configuration is generated by New-SwitchConfiguration.ps1 from tor.os10.template (toolkit registry 5.0.62); QoS follows Network ATC |
End of Task