RDG for DPF Zero Trust (DPF-ZT) with HBN DPU Service
Created on Sep 09, 2025 Updated on Jan 18, 2026 (v 25.10 GA) Scope This Reference Deployment Guide (RDG) provides comprehensive instructions for deploying the
文档目录
Created on Sep 09, 2025
Updated on Jan 18, 2026 (v 25.10 GA)
Scope
This Reference Deployment Guide (RDG) provides comprehensive instructions for deploying the NVIDIA DOCA Platform Framework (DPF) on high-performance, bare-metal infrastructure in Zero-Trust mode. The guide focuses on setting up an accelerated Host-Based Networking (HBN) service on NVIDIA® BlueField®-3 DPU to deliver secure, isolated, and hardware-accelerated environments.
The guide is intended for experienced system administrators, systems engineers, and solution architects who build highly secure bare-metal environments with Host-Based Networking enabled using NVIDIA BlueField DPU for acceleration, isolation, and infrastructure offload.
This document is an extension of the RDG for DPF Zero Trust (DPF-ZT) - NVIDIA Docs (referred to as the Baseline RDG). It details the additional steps and modifications required to deploy the HBN Service into the Baseline RDG environment.
Warning
- This reference implementation, as the name implies, is a specific, opinionated deployment example designed to address the use case described above.
- Although other approaches may exist for implementing similar solutions, this document provides a detailed guide for this specific method.
Abbreviations and Acronyms
| Term | Definition | Term | Definition |
|---|---|---|---|
| BFB | BlueField Bootstream | NFS | Network File System |
| BGP | Border Gateway Protocol | OOB | Out-of-Band |
| DOCA | Data Center Infrastructure-on-a-Chip Architecture | PF | Physical Function |
| DPF | DOCA Platform Framework | RDG | Reference Deployment Guide |
| DPU | Data Processing Unit | RDMA | Remote Direct Memory Access |
| HBN | Host Based Networking | RoCE | RDMA over Converged Ethernet |
| IPAM | IP Address Management | SFC | Service Function Chaining |
| K8S | Kubernetes | SR-IOV | Single Root Input/Output Virtualization |
| KVM | Kernel-based Virtual Machine | VLAN | Virtual LAN (Local Area Network) |
| MAAS | Metal as a Service | VNI | Virtual Network Interface |
| MTU | Maximum Transmission Unit | VRF | Virtual Router/Forwarder |
| NGC | NVIDIA GPU Cloud | ZT | Zero Trust |
Introduction
The NVIDIA BlueField-3 Data Processing Unit (DPU) is a 400 Gb/s infrastructure compute platform designed for line-rate processing of software-defined networking, storage, and cybersecurity workloads. It combines powerful compute resources, high-speed networking, and advanced programmability to deliver hardware-accelerated, software-defined solutions for modern data centers.
NVIDIA DOCA unleashes the full potential of the BlueField platform by enabling rapid development of applications and services that offload, accelerate, and isolate data center workloads.
One such service is Host-Based Networking (HBN) - a DOCA-enabled solution that allows network architects to design networks based on Layer 3 (L3) protocols. HBN enables routing on the server side by using BlueField as a BGP router. It encapsulates key networking functions in a containerized service pod, deployed directly on the BlueField’s Arm cores.
However, deploying and managing DPU and their associated DOCA services, especially at scale, presents operational challenges. Without a robust provisioning and orchestration system, tasks such as lifecycle management, service deployment, and network configuration for service function chaining (SFC) can quickly become complex and error prone. This is where the DOCA Platform Framework (DPF) comes into play.
DPF automates the full DPU lifecycle, streamlines the deployment of DOCA services, and simplifies advanced network configurations. With DPF, services such as HBN can be deployed seamlessly, allowing for efficient offloading and intelligent routing of traffic through the DPU data plane.
By leveraging DPF, users can scale and automate DPU management across Bare Metal, Virtual, and Kubernetes customer environments - optimizing performance while simplifying operations.
DPF supports multiple deployment models. This guide focuses on the Zero Trust bare-metal deployment model. In this scenario:
- The DPU is managed through its Baseboard Management Controller (BMC)
- All management traffic occurs over the DPU's out-of-band (OOB) network
- The host is considered as an untrusted entity towards the data center network. The DPU acts as a barrier between the host and the network.
- The host sees the DPU as a standard NIC, with no access to the internal DPU management plane (Zero Trust Mode)
This Reference Deployment Guide (RDG) provides a step-by-step example for installing DPF in Zero-Trust mode and HBN. It also includes practical demonstrations of performance optimization, validated using standard RDMA and TCP workloads.
As part of the reference implementation, open-source components outside the scope of DPF (e.g., MAAS, pfSense, Kubespray) are used to simulate a realistic customer deployment environment. The guide includes the full end-to-end deployment process, including:
- Infrastructure provisioning
- DPF deployment
- DPU provisioning (redfish)
- Service configuration and deployment
- Service chaining.
This document extends the capabilities of the DPF-managed Kubernetes cluster described in the RDG for DPF Zero Trust (DPF-ZT) - NVIDIA Docs (referred to as the Baseline RDG) by deploying the NVIDIA DOCA HBN Service within the existing DPF deployment to achieve a comprehensive, accelerated infrastructure.
References
Solution Architecture
Key Components and Technologies
-
NVIDIA BlueField® Data Processing Unit (DPU) The NVIDIA® BlueField® data processing unit (DPU) ignites unprecedented innovation for modern data centers and supercomputing clusters. With its robust compute power and integrated software-defined hardware accelerators for networking, storage, and security, BlueField creates a secure and accelerated infrastructure for any workload in any environment, ushering in a new era of accelerated computing and AI.
-
NVIDIA DOCA Software Framework NVIDIA DOCA™ unlocks the potential of the NVIDIA® BlueField® networking platform. By harnessing the power of BlueField DPU and SuperNICs, DOCA enables the rapid creation of applications and services that offload, accelerate, and isolate data center workloads. It lets developers create software-defined, cloud-native, DPU- and SuperNIC-accelerated services with zero-trust protection, addressing the performance and security demands of modern data centers.
-
NVIDIA ConnectX SmartNICs 10/25/40/50/100/200 and 400G Ethernet Network 网卡 The industry-leading NVIDIA® ConnectX® family of smart network interface cards (SmartNICs) offer advanced hardware offloads and accelerations. NVIDIA Ethernet adapters enable the highest ROI and lowest Total Cost of Ownership for hyperscale, public and private clouds, storage, machine learning, AI, big data, and telco platforms.
-
NVIDIA LinkX Cables The NVIDIA® LinkX® product family of cables and transceivers provides the industry’s most complete line of 10, 25, 40, 50, 100, 200, and 400GbE in Ethernet and 100, 200 and 400Gb/s InfiniBand products for Cloud, HPC, hyperscale, Enterprise, telco, storage and artificial intelligence, data center applications.
-
NVIDIA Spectrum 以太网交换机 Flexible form-factors with 16 to 128 physical ports, supporting 1GbE through 400GbE speeds. Based on a ground-breaking silicon technology optimized for performance and scalability, NVIDIA Spectrum switches are ideal for building high-performance, cost-effective, and efficient Cloud Data Center Networks, Ethernet Storage Fabric, and Deep Learning Interconnects. NVIDIA combines the benefits of NVIDIA Spectrum™ switches, based on an industry-leading application-specific integrated circuit (ASIC) technology, with a wide variety of modern network operating system choices, including NVIDIA Cumulus® Linux, SONiC and NVIDIA Onyx®.
-
NVIDIA Cumulus Linux NVIDIA® Cumulus® Linux is the industry's most innovative open network operating system that allows you to automate, customize, and scale your data center network like no other.
-
Kubernetes Kubernetes is an open-source container orchestration platform for deployment automation, scaling, and management of containerized applications.
-
Kubespray Kubespray is a composition of Ansible playbooks, inventory, provisioning tools, and domain knowledge for generic OS/Kubernetes clusters configuration management tasks and provides:
- A highly available cluster
- Composable attributes
- Support for most popular Linux distributions
Solution Design
Solution Logical Design
The logical design includes the following components:
- 1 x Hypervisor node (KVM-based) with ConnectX-7:
- 1 x Firewall VM
- 1 x Jump Node VM
- 1 x MaaS VM
- 3 x K8s Master VMs running all K8s management components
- 4 x Worker nodes (PCI Gen5), each with a 1 x BlueField-3 NIC
- Single High-Speed (HS) switch
- 1 Gb Host Management network
HBN service Logical Design
As part of this RDG, we will:
- Create two fully isolated logical networks per bare-metal workload server using a single physical function (PF0).
- Connect each network through the HBN service to a dedicated VLAN/VNI, mapped to separate VRFs (RED or BLUE).
- Route all workload traffic through the HBN service, with routing and isolation enforced inside the DPU.
- Assign PF0 as the sole network interface for each bare-metal workload server, with no networking configuration on the host.
- Demonstrate accelerated RDMA and TCP traffic between workload servers running on different bare-metal hosts within the same network (for example, RED ↔ RED).
- Validate strict network isolation by confirming that traffic between workloads in different networks (RED vs BLUE) is not permitted.
Firewall Design
The pfSense firewall in this solution serves a dual purpose:
- Firewall—provides an isolated environment for the DPF system, ensuring secure operations
- Router—enables Internet access for the management network
Port-forwarding rules for SSH and RDP are configured on the firewall to route traffic to the jump node’s IP address in the host management network. From the jump node, administrators can manage and access various devices in the setup, as well as handle the deployment of the Kubernetes (K8s) cluster and DPF components.
The following diagram illustrates the firewall design used in this solution:
Software Stack Components
注意： 确保使用上述完全相同的软件栈版本。
Bill of Materials
Deployment and Configuration
Node and Switch Definitions
以下是部署所演示网络结构时使用的定义和参数：
交换机 Ports Usage
| Hostname | Rack ID | Ports |
|---|---|---|
mgmt-switch |
1 | swp1-3 |
hs-switch |
1 | swp1-9 |
Hosts
| Rack | Server Type | Server Name | Switch Port | IP and NICs | Default Gateway |
|---|---|---|---|---|---|
| Rack1 | Hypervisor Node | hypervisor |
mgmt-switch: swp1hs-switch: swp1 |
lab-br (interface eno1): Trusted LAN IPmgmt-br (interface eno2): -hs-br (interface enp1s0): - | Trusted LAN GW |
| Rack1 | Firewall (Virtual) | fw |
- | WAN (lab-br): Trusted LAN IPLAN (mgmt-br): 10.0.110.254/24OPT1(hs-br): 10.0.123.254/22 | Trusted LAN GW |
| Rack1 | Jump Node (Virtual) | jump |
- | enp1s0: 10.0.110.253/24 | 10.0.110.254 |
| Rack1 | MaaS (Virtual) | maas |
- | enp1s0: 10.0.110.252/24 | 10.0.110.254 |
| Rack1 | Master Node (Virtual) | master1 |
- | enp1s0: 10.0.110.1/24 | 10.0.110.254 |
| Rack1 | Master Node (Virtual) | master2 |
- | enp1s0: 10.0.110.2/24 | 10.0.110.254 |
| Rack1 | Master Node (Virtual) | master3 |
- | enp1s0: 10.0.110.3/24 | 10.0.110.254 |
| Rack1 | Worker Node | worker1 |
mgmt-switch: swp2(DPU OOB)hs-switch: swp2-swp3 |
dpubmc: 10.0.110.21/24ens1f0np0/ens1f1np1: 10.0.120.0/22 | 10.0.110.254 |
| Rack1 | Worker Node | worker2 |
mgmt-switch: swp3(DPU OOB)hs-switch: swp4-swp5 |
dpubmc: 10.0.110.22/24ens1f0np0/ens1f1np1: 10.0.120.0/22 | 10.0.110.254 |
| Rack1 | Worker Node | worker3 |
mgmt-switch: swp2(DPU OOB)hs-switch: swp6-swp7 |
dpubmc: 10.0.110.23/24ens1f0np0/ens1f1np1: 10.0.120.0/22 | 10.0.110.254 |
| Rack1 | Worker Node | worker4 |
mgmt-switch: swp3(DPU OOB)hs-switch: swp8-swp9 |
dpubmc: 10.0.110.24/24ens1f0np0/ens1f1np1: 10.0.120.0/22 | 10.0.110.254 |
style="--text-align: center">ens1f0np0/ens1f1np1: 10.0.120.0/22 10.0.110.254
虚拟机监控程序节点
裸金属工作节点
网络结构配置
更新 Cumulus Linux
作为最佳实践，请确保使用最新发布的 Cumulus Linux NOS 版本。
有关如何升级 Cumulus Linux 的信息，请参阅 Cumulus Linux 用户指南。
配置 Cumulus Linux 交换机
SN3700 交换机 (hs-switch) 配置如下：
nv set evpn state enable
nv set interface eth0 ip address dhcp
nv set interface eth0 ip vrf mgmt
nv set interface eth0 type eth
nv set interface lo ipv4 address 11.0.0.101/32
nv set interface lo type loopback
nv set interface swp1-17 link state up
nv set interface swp1-17 type swp
nv set interface swp1 ipv4 address 10.0.123.253/22
nv set router bgp autonomous-system 65001
nv set router bgp state enabled
nv set router bgp graceful-restart mode full
nv set router bgp router-id 11.0.0.101
nv set vrf default router bgp address-family ipv4-unicast state enabled
nv set vrf default router bgp address-family ipv4-unicast redistribute connected state enabled
nv set vrf default router bgp address-family ipv4-unicast redistribute static state enabled
nv set vrf default router bgp address-family ipv6-unicast state enabled
nv set vrf default router bgp address-family ipv6-unicast redistribute connected state enabled
nv set vrf default router bgp address-family l2vpn-evpn state enabled
nv set vrf default router bgp state enabled
nv set vrf default router bgp neighbor swp2 peer-group hbn
nv set vrf default router bgp neighbor swp2 type unnumbered
nv set vrf default router bgp neighbor swp3 peer-group hbn
nv set vrf default router bgp neighbor swp3 type unnumbered
nv set vrf default router bgp neighbor swp4 peer-group hbn
nv set vrf default router bgp neighbor swp4 type unnumbered
nv set vrf default router bgp neighbor swp5 peer-group hbn
nv set vrf default router bgp neighbor swp5 type unnumbered
nv set vrf default router bgp neighbor swp6 peer-group hbn
nv set vrf default router bgp neighbor swp6 type unnumbered
nv set vrf default router bgp neighbor swp7 peer-group hbn
nv set vrf default router bgp neighbor swp7 type unnumbered
nv set vrf default router bgp neighbor swp8 peer-group hbn
nv set vrf default router bgp neighbor swp8 type unnumbered
nv set vrf default router bgp neighbor swp9 peer-group hbn
nv set vrf default router bgp neighbor swp9 type unnumbered
nv set vrf default router bgp path-selection multipath aspath-ignore enabled
nv set vrf default router bgp peer-group hbn address-family l2vpn-evpn state enabled
nv set vrf default router bgp peer-group hbn remote-as external
nv set vrf default router static 0.0.0.0/0 address-family ipv4-unicast
nv set vrf default router static 0.0.0.0/0 via 10.0.123.254 type ipv4-address
nv config apply -y
nv config save
SN2201 交换机 (mgmt-switch) 配置如下：
nv set interface swp1-3 link state up
nv set interface swp1-3 type swp
nv set interface swp1-3 bridge domain br_default
nv set bridge domain br_default untagged 1
nv config apply
nv config save -y
主机配置
注意： 确保工作节点服务器的 BIOS 设置已启用 SR-IOV，并且服务器已调整为最大性能。
所有工作节点必须具有相同的 BlueField-3 网卡 PCIe 位置，并且必须显示相同的接口名称。
确保您拥有 DPU BMC 和 OOB MAC 地址。
与参考部署指南（基线 RDG）（“部署和配置”部分，“主机配置”小节）相比无变化。
虚拟机监控程序安装和配置
与基线 RDG（“部署和配置”部分，“虚拟机监控程序安装和配置”小节）相比无变化。
准备基础设施服务器
关于防火墙 VM、Jump VM、MaaS VM，与基线 RDG（“部署和配置”部分，“准备基础设施服务器”小节）相比无变化。
（可选）防火墙 VM – 裸金属服务器外部连接
为了通过高速网络提供从裸金属主机到外部的连接，打开 Firefox 浏览器并访问 pfSense Web UI（http://10.0.110.254）。
- 系统：
-
路由 → 网关 → 添加 → “接口”：OPT1，“地址族”：IPv4，“名称”：switch，“网关”：10.0.123.253 → 点击“保存”→ 在“默认网关” - “默认网关 IPv4”下选择 WAN_DHCP → 点击“保存”
警告： 请注意，“网关”和“监控 IP”下的 Trusted LAN 网络 IP 地址已模糊处理。
-
使用 MaaS 配置主 VM
与基线 RDG（“部署和配置”部分，“使用 MaaS 配置主 VM”小节）相比无变化。
K8s 集群部署和配置
初始 Kubernetes 集群部署（使用 Kubespray 部署主节点）及后续验证的步骤与基线 RDG（“K8s 集群部署和配置”部分，子节：“Kubespray 部署和配置”、“使用 Kubespray Ansible Playbook 部署集群”、“K8s 部署验证”）相比保持不变。
DPF 安装
DPF 安装过程（Operator、系统组件）基本遵循基线 RDG。
软件前提条件和所需变量
-
首先安装剩余的软件前提条件。
Jump 节点控制台
## 连接到 master1 以复制 kubespray 部署期间安装的 helm 客户端工具 $ depuser@jump:~$ ssh master1 depuser@master1:~$ cp /usr/local/bin/helm /tmp/ ## 在另一个标签页中 depuser@jump:~$ scp master1:/tmp/helm /tmp/ depuser@jump:~$ sudo chown root:root /tmp/helm depuser@jump:~$ sudo mv /tmp/helm /usr/local/bin/ ## 验证 envsubst 工具是否已安装 depuser@jump:~$ which envsubst /usr/bin/envsubst -
继续克隆 doca-platform Git 仓库：
Jump 节点控制台
$ git clone https://github.com/NVIDIA/doca-platform.git
-
Change directory to doca-platform and checkout to tag v25.10.0:
Jump Node Console
$ cd doca-platform/ $ git checkout v25.10.0 -
Change directory to readme.md from where all the commands will be run:
Jump Node Console
$ cd doca-platform/docs/public/user-guides/zero-trust/use-cases/hbn -
Change the BMC root's password. In Zero Trust mode, provisioning DPU requires authentication with Redfish. In order to do that, you must set the same root password to access the BMC for all DPU DPF is going to manage. For more information on how to set the BMC root password refer to BlueField DPU Administrator Quick Start Guide.
Connect to the first DPU BMC over SSH to change the BMC root's password:
Jump Node Console
$ ssh root@10.0.110.201 root@10.0.110.201's password: <BMC Root Password. Default root/0penBmc. need to change first time to $BMC_ROOT_PASSWORD in the manifests/00-env-vars/envvars.env file> -
Modify the variables in
manifests/00-env-vars/envvars.envto fit your environment, then source the file:Warning Replace the values for the variables in the following file with the values that fit your setup. Specifically, pay attention to
DPUCLUSTER_INTERFACE,BMC_ROOT_PASSWORD, andDPU's serial number. To get aDPU's serial numberyou can use following command. Sample: $ curl -k -u root:'BMC root password' https://10.0.110.201/redfish/v1/Systems/Bluefield | jq -r '.SerialNumber | ascii_downcase' % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 100 4970 100 4970 0 0 4211 0 0:00:01 0:00:01 --:--:-- 4211 mt2402xz0f7xmanifests/00-env-vars/envvars.env
## IP Address for the Kubernetes API server of the target cluster on which DPF is installed. ## This should never include a scheme or a port. ## e.g. 10.10.10.10 export TARGETCLUSTER_API_SERVER_HOST=10.0.110.10 ## Virtual IP used by the load balancer for the DPU Cluster. Must be a reserved IP from the management subnet and not ## allocated by DHCP. export DPUCLUSTER_VIP=10.0.110.200 ## Interface on which the DPUCluster load balancer will listen. Should be the management interface of the control plane node. export DPUCLUSTER_INTERFACE=ens160 ## IP address to the NFS server used as storage for the BFB. export NFS_SERVER_IP=10.0.110.253 ## The repository URL for the NVIDIA Helm chart registry. ## Usually this is the NVIDIA Helm NGC registry. For development purposes, this can be set to a different repository. export HELM_REGISTRY_REPO_URL=https://helm.ngc.nvidia.com/nvidia/doca ## The repository URL for the HBN container image. ## Usually this is the NVIDIA NGC registry. For development purposes, this can be set to a different repository. export HBN_NGC_IMAGE_URL=nvcr.io/nvidia/doca/doca_hbn ## The DPF REGISTRY is the Helm repository URL where the DPF Operator Chart resides. ## Usually this is the NVIDIA Helm NGC registry. For development purposes, this can be set to a different repository. export REGISTRY=https://helm.ngc.nvidia.com/nvidia/doca ## The DPF TAG is the version of the DPF components which will be deployed in this guide. export TAG=v25.10.0 ## URL to the BFB used in the `bfb.yaml` and linked by the DPUSet. export BFB_URL="https://content.mellanox.com/BlueField/BFBs/Ubuntu24.04/bf-bundle-3.2.1-34_25.11_ubuntu-24.04_64k_prod.bfb" ## IP_RANGE_START and IP_RANGE_END ## These define the IP range for DPU discovery via Redfish/BMC interfaces ## Example: If your DPU have BMC IPs in range 10.0.110.201-224 ## export IP_RANGE_START=10.0.110.201 ## export IP_RANGE_END=10.0.110.224 ## Start of DPUDiscovery IpRange export IP_RANGE_START=10.0.110.201 ## End of DPUDiscovery IpRange export IP_RANGE_END=10.0.110.204 # The password used for DPU BMC root login, must be the same for all DPU # For more information on how to set the BMC root password refer to BlueField DPU Administrator Quick Start Guide. export BMC_ROOT_PASSWORD=<set your BMC_ROOT_PASSWORD> ## Serial number of DPU. If you have more than 2 DPU, you will need to parameterize the system accordingly and expose ## additional variables. ## All serial numbers must be in lowercase. ## Serial number of DPU1 export DPU1_SERIAL=mt2402xz0f7x ## Serial number of DPU2 export DPU2_SERIAL=mt2402xz0f80 ## Serial number of DPU3 export DPU2_SERIAL=mt2402xz0f9n ## Serial number of DPU4 export DPU2_SERIAL=mt2402xz0f8g -
Export environment variables for the installation:
Jump Node Console
$ source manifests/00-env-vars/envvars.env
DPF Operator Installation
No change from the Baseline RDG (Section "DPF Installation", Subsection "DPF Operator Installation").
DPF System Installation
No change from the Baseline RDG (Section "DPF Installation", Subsection "DPF System Installation").
DPU Services Installation
HBN DPU Service Installation
This section focuses on provisioning NVIDIA®BlueField®-3 DPU using DPF, installing the HBN DPU Service on those DPU and enabling workload traffic to pass through HBN before leaving the DPU.
-
Export environment variables for the installation:
Jump Node Console
$ source manifests/00-env-vars/envvars.env -
Use the following YAML to define a
BFBresource that downloads the Bluefield Bitstream to a shared volume:--- apiVersion: provisioning.dpu.nvidia.com/v1alpha1 kind: BFB metadata: name: bf-bundle-$TAG namespace: dpf-operator-system spec: url: $BFB_URL -
Change the DPUFlavor using the following YAML:
--- apiVersion: provisioning.dpu.nvidia.com/v1alpha1 kind: DPUFlavor metadata: name: hbn-$TAG namespace: dpf-operator-system spec: dpuMode: zero-trust bfcfgParameters: - UPDATE_ATF_UEFI=yes - UPDATE_DPU_OS=yes - WITH_NIC_FW_UPDATE=yes configFiles: - operation: override path: /etc/mellanox/mlnx-bf.conf permissions: "0644" raw: | ALLOW_SHARED_RQ="no" IPSEC_FULL_OFFLOAD="no" ENABLE_ESWITCH_MULTIPORT="yes" - operation: override path: /etc/mellanox/mlnx-ovs.conf permissions: "0644" raw: | CREATE_OVS_BRIDGES="no" OVS_DOCA="yes" - operation: override path: /etc/mellanox/mlnx-sf.conf permissions: "0644" raw: "" grub: kernelParameters: - console=hvc0 - console=ttyAMA0 - earlycon=pl011,0x13010000 - fixrttc - net.ifnames=0 - biosdevname=0 - iommu.passthrough=1 - cgroup_no_v1=net_prio,net_cls - hugepagesz=2048kB - hugepages=3072 nvconfig: - device: '*' parameters: - PF_BAR2_ENABLE=0 - PER_PF_NUM_SF=1 - PF_TOTAL_SF=20 - PF_SF_BAR_SIZE=10 - NUM_PF_MSIX_VALID=0 - PF_NUM_PF_MSIX_VALID=1 - PF_NUM_PF_MSIX=228 - INTERNAL_CPU_MODEL=1 - INTERNAL_CPU_OFFLOAD_ENGINE=0 - SRIOV_EN=1 - NUM_OF_VFS=46 - LAG_RESOURCE_ALLOCATION=1 - LINK_TYPE_P1=ETH - LINK_TYPE_P2=ETH - EXP_ROM_UEFI_x86_ENABLE=1 ovs: rawConfigScript: | _ovs-vsctl() { ovs-vsctl --no-wait --timeout 15 "$@" } _ovs-vsctl set Open_vSwitch . other_config:doca-init=true _ovs-vsctl set Open_vSwitch . other_config:dpdk-max-memzones=50000 _ovs-vsctl set Open_vSwitch . other_config:hw-offload=true _ovs-vsctl set Open_vSwitch . other_config:pmd-quiet-idle=true _ovs-vsctl set Open_vSwitch . other_config:max-idle=20000 _ovs-vsctl set Open_vSwitch . other_config:max-revalidator=5000 _ovs-vsctl --if-exists del-br ovsbr1 _ovs-vsctl --if-exists del-br ovsbr2 _ovs-vsctl --may-exist add-br br-sfc _ovs-vsctl set bridge br-sfc datapath_type=netdev _ovs-vsctl set bridge br-sfc fail_mode=secure _ovs-vsctl --may-exist add-port br-sfc p0 _ovs-vsctl set Interface p0 type=dpdk _ovs-vsctl set Interface p0 mtu_request=9216 _ovs-vsctl set Port p0 external_ids:dpf-type=physical
_ovs-vsctl --may-exist add-port br-sfc p1
_ovs-vsctl set Interface p1 type=dpdk
_ovs-vsctl set Interface p1 mtu_request=9216
_ovs-vsctl set Port p1 external_ids:dpf-type=physical
_ovs-vsctl --may-exist add-br br-hbn
_ovs-vsctl set bridge br-hbn datapath_type=netdev
_ovs-vsctl set bridge br-hbn fail_mode=secure
-
Change the
dpudeployment.yamlfile to reference the DPUFlavor.--- apiVersion: svc.dpu.nvidia.com/v1alpha1 kind: DPUDeployment metadata: name: hbn-only namespace: dpf-operator-system spec: dpus: bfb: bf-bundle-$TAG flavor: hbn-$TAG nodeEffect: hold: true dpuSets: - nameSuffix: "dpuset1" nodeSelector: matchLabels: feature.node.kubernetes.io/dpu-enabled: "true" services: doca-hbn: serviceTemplate: doca-hbn serviceConfiguration: doca-hbn serviceChains: switches: - ports: - serviceInterface: matchLabels: uplink: p0 - service: name: doca-hbn interface: p0_if - ports: - serviceInterface: matchLabels: uplink: p1 - service: name: doca-hbn interface: p1_if - ports: - serviceInterface: matchLabels: interface: pf0hpf - service: interface: pf0hpf_if name: doca-hbnWarning: Please notice that with default nodeEffect above, DPU provisioning workflow will be paused and wait for an external signal (annotation) in order to proceed, as demonstrated in upcoming steps. To implement a fully automated process that won’t require user intervention, see customAction option.
-
Change the rest of the configuration files.
As explained in the introduction, these files create service chains that connect two physical functions PF0 or PF0 to the outer fabric through HBN, providing EVPN VXLAN overlay, VNI based isolation, and ECMP redundancy across both DPU uplinks (p0 and p1). These are the configuration files.
-
HBN DPUServiceConfig and DPUServiceTemplate to deploy HBN workloads to the DPU.
--- apiVersion: svc.dpu.nvidia.com/v1alpha1 kind: DPUServiceConfiguration metadata: name: doca-hbn namespace: dpf-operator-system spec: deploymentServiceName: "doca-hbn" serviceConfiguration: serviceDaemonSet: annotations: k8s.v1.cni.cncf.io/networks: |- [ {"name": "iprequest", "interface": "ip_lo", "cni-args": {"poolNames": ["loopback"], "poolType": "cidrpool"}}, {"name": "iprequest", "interface": "ip_pf0hpf_red", "cni-args": {"poolNames": ["pool1"], "poolType": "cidrpool", "allocateDefaultGateway": true}}, {"name": "iprequest", "interface": "ip_pf0hpf_blue", "cni-args": {"poolNames": ["pool2"], "poolType": "cidrpool", "allocateDefaultGateway": true}} ] helmChart: values: configuration: perDPUValuesYAML: | - hostnamePattern: "*" values: bgp_peer_group: hbn # ---- DPU1, DPU2 => RED only ---- - hostnamePattern: "dpu-node-mt2402xz0f7x-mt2402xz0f7x*" values: role: RED vrf: RED vlan: 11 l2vni: 10010 l3vni: 100001 bgp_autonomous_system: 65101 - hostnamePattern: "dpu-node-mt2402xz0f80-mt2402xz0f80*" values: role: RED vrf: RED vlan: 11 l2vni: 10010 l3vni: 100001 bgp_autonomous_system: 65201 # ---- DPU3, DPU4 => BLUE only ---- - hostnamePattern: "dpu-node-mt2402xz0f9n-mt2402xz0f9n*" values: role: BLUE vrf: BLUE vlan: 21 l2vni: 10020 l3vni: 100002 bgp_autonomous_system: 65301 - hostnamePattern: "dpu-node-mt2402xz0f8g-mt2402xz0f8g*" values: role: BLUE vrf: BLUE vlan: 21 l2vni: 10020 l3vni: 100002 bgp_autonomous_system: 65401 startupYAMLJ2: | - header: model: bluefield nvue-api-version: nvue_v1 rev-id: 1.0 version: HBN 2.4.0 - set: bridge: domain: br_default: vlan: {{ config.vlan }}: vni: {{ config.l2vni }}: {} evpn: enable: on route-advertise: {} interface: lo: ip: address: {{ ipaddresses.ip_lo.ip }}/32: {} type: loopback p0_if,p1_if,pf0hpf_if: type: swp link: mtu: 9000 pf0hpf_if: bridge: domain: br_default: access: {{ config.vlan }} vlan{{ config.vlan }}: type: svi vlan: {{ config.vlan }} ip: address: {% if config.role == "RED" %} {{ ipaddresses.ip_pf0hpf_red.cidr }}: {} {% else %} {{ ipaddresses.ip_pf0hpf_blue.cidr }}: {} {% endif %} vrf: {{ config.vrf }} nve: vxlan: arp-nd-suppress: on enable: on source: address: {{ ipaddresses.ip_lo.ip }} router: bgp: enable: on graceful-restart: mode: full vrf: default: router: bgp: address-family: ipv4-unicast: enable: on redistribute: connected: enable: on l2vpn-evpn: enable: on autonomous-system: {{ config.bgp_autonomous_system }} enable: on neighbor: p0_if: peer-group: {{ config.bgp_peer_group }} type: unnumbered p1_if: peer-group: {{ config.bgp_peer_group }} type: unnumbered path-selection: multipath: aspath-ignore: on peer-group: {{ config.bgp_peer_group }}: address-family: ipv4-unicast: enable: on l2vpn-evpn: enable: on remote-as: external router-id: {{ ipaddresses.ip_lo.ip }} {{ config.vrf }}: evpn: enable: on vni: {{ config.l3vni }}: {} loopback: ip: address: {{ ipaddresses.ip_lo.ip }}/32: {} router: bgp: address-family: ipv4-unicast: enable: on redistribute: connected: enable: on route-export: to-evpn: enable: on autonomous-system: {{ config.bgp_autonomous_system }} enable: on router-id: {{ ipaddresses.ip_lo.ip }} interfaces: - name: p0_if network: mybrhbn - name: p1_if network: mybrhbn - name: pf0hpf_if network: mybrhbn--- apiVersion: svc.dpu.nvidia.com/v1alpha1 kind: DPUServiceTemplate metadata: name: doca-hbn namespace: dpf-operator-system spec: deploymentServiceName: "doca-hbn" helmChart: source: repoURL: $HELM_REGISTRY_REPO_URL version: 1.0.5 chart: doca-hbn values: image: repository: $HBN_NGC_IMAGE_URL tag: 3.2.1-doca3.2.1 resources: memory: 6Gi nvidia.com/bf_sf: 4 -
Physical Interfaces for physical ports on the DPU.
--- apiVersion: svc.dpu.nvidia.com/v1alpha1 kind: DPUServiceInterface metadata: name: p0 namespace: dpf-operator-system spec: template: spec: template: metadata: labels: uplink: "p0" spec: interfaceType: physical physical: interfaceName: p0 --- apiVersion: svc.dpu.nvidia.com/v1alpha1 kind: DPUServiceInterface metadata: name: p1 namespace: dpf-operator-system spec: template: spec: template: metadata: labels: uplink: "p1" spec: interfaceType: physical physical: interfaceName: p1 --- apiVersion: svc.dpu.nvidia.com/v1alpha1 kind: DPUServiceInterface metadata: name: pf0hpf namespace: dpf-operator-system spec: template: spec: template: metadata: labels: interface: "pf0hpf" spec: interfaceType: pf pf: pfID: 0 -
DPU Service IPAM objects to set up IP Address Management on the DPUCluster.
--- apiVersion: svc.dpu.nvidia.com/v1alpha1 kind: DPUServiceIPAM metadata: name: pool1 namespace: dpf-operator-system spec: ipv4Network: network: "10.0.121.0/24" gatewayIndex: 2 prefixSize: 29
-
These preallocations are not necessary. We specify them so that the validation commands are straightforward.
allocations:
dpu-node-${DPU1_SERIAL}-${DPU1_SERIAL}: 10.0.121.0/29 dpu-node-${DPU2_SERIAL}-${DPU2_SERIAL}: 10.0.121.8/29
apiVersion: svc.dpu.nvidia.com/v1alpha1 kind: DPUServiceIPAM metadata: name: pool2 namespace: dpf-operator-system spec: ipv4Network: network: "10.0.122.0/24" gatewayIndex: 2 prefixSize: 29 allocations: dpu-node-${DPU3_SERIAL}-${DPU3_SERIAL}: 10.0.122.0/29 dpu-node-${DPU4_SERIAL}-${DPU4_SERIAL}: 10.0.122.8/29
apiVersion: svc.dpu.nvidia.com/v1alpha1 kind: DPUServiceIPAM metadata: name: loopback namespace: dpf-operator-system spec: ipv4Network: network: "11.0.0.0/24" prefixSize: 32
Warning It is necessary to set several environment variables before running this command.
$ source manifests/00-env-vars/envvars.env
-
Apply all of the YAML files mentioned above using the following command:
Jump Node Console
$ cat manifests/03.1-dpudeployment-installation-pf/*.yaml | envsubst | kubectl apply -f -Jump Node Console
$ kubectl wait --for=condition=ApplicationsReconciled --namespace dpf-operator-system dpuservices --all dpuservice.svc.dpu.nvidia.com/doca-hbn-wb5pg condition met dpuservice.svc.dpu.nvidia.com/flannel condition met dpuservice.svc.dpu.nvidia.com/multus condition met dpuservice.svc.dpu.nvidia.com/nvidia-k8s-ipam condition met dpuservice.svc.dpu.nvidia.com/ovs-cni condition met dpuservice.svc.dpu.nvidia.com/servicechainset-controller condition met dpuservice.svc.dpu.nvidia.com/servicechainset-rbac-and-crds condition met dpuservice.svc.dpu.nvidia.com/sfc-controller condition met dpuservice.svc.dpu.nvidia.com/sriov-device-plugin condition met $ kubectl wait --for=condition=DPUIPAMObjectReconciled --namespace dpf-operator-system dpuserviceipam --all dpuserviceipam.svc.dpu.nvidia.com/loopback condition met dpuserviceipam.svc.dpu.nvidia.com/pool1 condition met dpuserviceipam.svc.dpu.nvidia.com/pool2 condition met $ kubectl wait --for=condition=ServiceInterfaceSetReconciled --namespace dpf-operator-system dpuserviceinterface --all dpuserviceinterface.svc.dpu.nvidia.com/doca-hbn-p0-if-vjqn5 condition met dpuserviceinterface.svc.dpu.nvidia.com/doca-hbn-p1-if-nl8rj condition met dpuserviceinterface.svc.dpu.nvidia.com/doca-hbn-pf0hpf-if-kbfj4 condition met dpuserviceinterface.svc.dpu.nvidia.com/doca-hbn-pf1hpf-if-79zsq condition met dpuserviceinterface.svc.dpu.nvidia.com/p0 condition met dpuserviceinterface.svc.dpu.nvidia.com/p1 condition met dpuserviceinterface.svc.dpu.nvidia.com/pf0hpf condition met dpuserviceinterface.svc.dpu.nvidia.com/pf1hpf condition met $ kubectl wait --for=condition=ServiceChainSetReconciled --namespace dpf-operator-system dpuservicechain --all dpuservicechain.svc.dpu.nvidia.com/hbn-only-8xrrx condition met -
To follow the progress of DPU provisioning, run the following command to check its current phase:Jump Node Console
$ watch -n10 "kubectl describe dpu -n dpf-operator-system | grep 'Node Name\|Type\|Last\|Phase'" -
Wait for the NodeEffect stage (at this point the provisioning is paused, waiting for external signal). Run following command on all/specific DPU nodemaintenance object/s to proceed with provisioning:
Jump Node Console
$ kubectl annotate dpunodemaintenances -n dpf-operator-system --all provisioning.dpu.nvidia.com/wait-for-external-nodeeffect=false --overwrite -
To follow the progress of DPU provisioning, run the following command to check its current phase:Jump Node Console
$ watch -n10 "kubectl describe dpu -n dpf-operator-system | grep 'Node Name\|Type\|Last\|Phase'" Every 10.0s: kubectl describe dpu -n dpf-operator-system | grep 'Node Name\|Type\|Last\|Phase' setup5-jump: Wed Jan 7 10:47:25 2026 Dpu Node Name: dpu-node-mt2402xz0f7x Last Transition Time: 2026-01-07T08:31:53Z Type: BFBPrepared Last Transition Time: 2026-01-07T08:31:49Z Type: BFBReady Last Transition Time: 2026-01-07T08:36:38Z Type: BFBTransferred Last Transition Time: 2026-01-07T08:31:52Z Type: FWConfigured Last Transition Time: 2026-01-07T08:31:49Z Type: Initialized Last Transition Time: 2026-01-07T08:31:50Z Type: InterfaceInitialized Last Transition Time: 2026-01-07T08:31:49Z Type: NodeEffectReady Last Transition Time: 2026-01-07T08:43:33Z Reason: OemLastState Type: OSInstalled Last Transition Time: 2026-01-07T08:46:37Z Type: Rebooted Phase: Rebooting Dpu Node Name: dpu-node-mt2402xz0f80 Last Transition Time: 2026-01-07T08:31:52Z Type: BFBPrepared Last Transition Time: 2026-01-07T08:31:49Z Type: BFBReady Last Transition Time: 2026-01-07T08:36:33Z Type: BFBTransferred Last Transition Time: 2026-01-07T08:31:51Z Type: FWConfigured Last Transition Time: 2026-01-07T08:31:49Z Type: Initialized Last Transition Time: 2026-01-07T08:31:49Z Type: InterfaceInitialized Last Transition Time: 2026-01-07T08:31:49Z Type: NodeEffectReady Last Transition Time: 2026-01-07T08:43:19Z Reason: OemLastState Type: OSInstalled Last Transition Time: 2026-01-07T08:46:23Z Type: Rebooted Phase: Rebooting ... -
Wait for the Rebooted stage and then Power Cycle the bare-metal host manually. After the DPU is up, run following command for each DPU worker:
Jump Node Console
$ kubectl -n dpf-operator-system annotate dpunode dpu-node-mt2402xz0f7x dpu-node-mt2402xz0f80 dpu-node-mt2402xz0f9n dpu-node-mt2402xz0f8g provisioning.dpu.nvidia.com/dpunode-external-reboot-required- -
At this point, the DPU workers should be added to the cluster. As they are being added to the cluster, the DPU are provisioned.
Jump Node Console
$ watch -n10 "kubectl describe dpu -n dpf-operator-system | grep 'Node Name\|Type\|Last\|Phase'" Every 10.0s: kubectl describe dpu -n dpf-operator-system | grep 'Node Name\|Type\|Last\|Phase' setup5-jump: Wed Jan 7 11:10:49 2026 Dpu Node Name: dpu-node-mt2402xz0f7x Type: InternalIP Type: Hostname Last Transition Time: 2026-01-07T09:09:57Z Type: Ready Last Transition Time: 2026-01-07T08:31:53Z Type: BFBPrepared Last Transition Time: 2026-01-07T08:31:49Z Type: BFBReady Last Transition Time: 2026-01-07T08:36:38Z Type: BFBTransferred Last Transition Time: 2026-01-07T09:09:57Z Type: DPUClusterReady Last Transition Time: 2026-01-07T08:31:52Z Type: FWConfigured Last Transition Time: 2026-01-07T08:31:49Z Type: Initialized Last Transition Time: 2026-01-07T08:31:50Z Type: InterfaceInitialized Last Transition Time: 2026-01-07T08:31:49Z Type: NodeEffectReady Last Transition Time: 2026-01-07T09:09:57Z Type: NodeEffectRemoved Last Transition Time: 2026-01-07T08:43:33Z Reason: OemLastState Type: OSInstalled Last Transition Time: 2026-01-07T09:09:57Z Type: Rebooted Phase: Ready Dpu Node Name: dpu-node-mt2402xz0f80 Type: InternalIP Type: Hostname Last Transition Time: 2026-01-07T09:10:24Z Type: Ready Last Transition Time: 2026-01-07T08:31:52Z Type: BFBPrepared Last Transition Time: 2026-01-07T08:31:49Z Type: BFBReady Last Transition Time: 2026-01-07T08:36:33Z Type: BFBTransferred Last Transition Time: 2026-01-07T09:10:24Z Type: DPUClusterReady Last Transition Time: 2026-01-07T08:31:51Z Type: FWConfigured Last Transition Time: 2026-01-07T08:31:49Z Type: Initialized Last Transition Time: 2026-01-07T08:31:49Z Type: InterfaceInitialized Last Transition Time: 2026-01-07T08:31:49Z Type: NodeEffectReady Last Transition Time: 2026-01-07T09:10:24Z Type: NodeEffectRemoved Last Transition Time: 2026-01-07T08:43:19Z Reason: OemLastState Type: OSInstalled Last Transition Time: 2026-01-07T09:10:24Z Type: Rebooted Phase: Ready ... -
Finally, validate that all the different DPU-related objects are
now in the Ready state:
Jump Node Console
$ kubectl get secrets -n dpu-cplane-tenant1 dpu-cplane-tenant1-admin-kubeconfig -o json | jq -r '.data["admin.conf"]' | base64 --decode > /home/depuser/dpu-cluster.config
$ echo "alias ki='KUBECONFIG=/home/depuser/dpu-cluster.config kubectl'" >> ~/.bashrc
$ echo 'alias dpfctl="kubectl -n dpf-operator-system exec deploy/dpf-operator-controller-manager -- /dpfctl "' >> ~/.bashrc
$ dpfctl describe dpudeployments
NAME NAMESPACE STATUS REASON SINCE MESSAGE
DPFOperatorConfig/dpfoperatorconfig dpf-operator-system Ready: True Success 3m3s
└─DPUDeployments
└─DPUDeployment/hbn dpf-operator-system Ready: True Success 22s
├─DPUServiceChains
│ └─DPUServiceChain/hbn-wd7fs dpf-operator-system Ready: True Success 65s
├─DPUServiceInterfaces
│ └─3 DPUServiceInterfaces... dpf-operator-system Ready: True Success 70s See doca-hbn-p0-if-749n9, doca-hbn-p1-if-fn8w5, doca-hbn-pf0hpf-if-9s8c6
├─DPUSets
│ └─DPUSet/hbn-dpuset1 dpf-operator-system Ready: True Success 71s
│ ├─BFB/bf-bundle-v25.10.0 dpf-operator-system Ready: True Ready 39m File: bf-bundle-3.2.1-34_25.11_ubuntu-24.04_64k_prod.bfb, DOCA: 3.2.1
│ ├─DPUNodes
│ │ └─4 DPUNodes... dpf-operator-system Ready: True Ready 98s See dpu-node-mt2402xz0f7x, dpu-node-mt2402xz0f80, dpu-node-mt2402xz0f8g, dpu-node-mt2402xz0f9n
│ └─DPU
│ └─4 DPU... dpf-operator-system Ready: True DPUReady 98s See dpu-node-mt2402xz0f7x-mt2402xz0f7x, dpu-node-mt2402xz0f80-mt2402xz0f80,
│ dpu-node-mt2402xz0f8g-mt2402xz0f8g, dpu-node-mt2402xz0f9n-mt2402xz0f9n
└─Services
├─DPUServiceTemplates
│ └─DPUServiceTemplate/doca-hbn dpf-operator-system Ready: True Success 39m
└─DPUServices
└─1 DPUServices... dpf-operator-system Ready: True Success 50s See doca-hbn-jxkxw
$ ki get node -A
NAME STATUS ROLES AGE VERSION
dpu-node-mt2402xz0f7x-mt2402xz0f7x Ready <none> 5m18s v1.34.3
dpu-node-mt2402xz0f80-mt2402xz0f80 Ready <none> 6m12s v1.34.3
dpu-node-mt2402xz0f8g-mt2402xz0f8g Ready <none> 6m14s v1.34.3
dpu-node-mt2402xz0f9n-mt2402xz0f9n Ready <none> 6m22s v1.34.3
$ kubectl get dpu -A
NAMESPACE NAME READY PHASE AGE
dpf-operator-system dpu-node-mt2402xz0f7x-mt2402xz0f7x True Ready 36m
dpf-operator-system dpu-node-mt2402xz0f80-mt2402xz0f80 True Ready 36m
dpf-operator-system dpu-node-mt2402xz0f8g-mt2402xz0f8g True Ready 36m
dpf-operator-system dpu-node-mt2402xz0f9n-mt2402xz0f9n True Ready 36m
$ kubectl wait --for=condition=ready --namespace dpf-operator-system dpu --all
dpu.provisioning.dpu.nvidia.com/dpu-node-mt2402xz0f7x-mt2402xz0f7x condition met
dpu.provisioning.dpu.nvidia.com/dpu-node-mt2402xz0f80-mt2402xz0f80 condition met
dpu.provisioning.dpu.nvidia.com/dpu-node-mt2402xz0f8g-mt2402xz0f8g condition met
dpu.provisioning.dpu.nvidia.com/dpu-node-mt2402xz0f9n-mt2402xz0f9n condition met
$ ki get pods -A -o wide
NAMESPACE NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
dpf-operator-system dpu-cplane-tenant1-cni-installer-89kn4 1/1 Running 0 6m50s 10.244.2.3 dpu-node-mt2402xz0f80-mt2402xz0f80 <none> <none>
dpf-operator-system dpu-cplane-tenant1-cni-installer-s8h4z 1/1 Running 0 7m1s 10.244.0.5 dpu-node-mt2402xz0f9n-mt2402xz0f9n <none> <none>
dpf-operator-system dpu-cplane-tenant1-cni-installer-wb29j 1/1 Running 0 5m57s 10.244.3.2 dpu-node-mt2402xz0f7x-mt2402xz0f7x <none> <none>
dpf-operator-system dpu-cplane-tenant1-cni-installer-zhzqh 1/1 Running 0 6m53s 10.244.1.4 dpu-node-mt2402xz0f8g-mt2402xz0f8g <none> <none>
dpf-operator-system dpu-cplane-tenant1-doca-hbn-jxkxw-ds-5sbzs 2/2 Running 0 2m54s 10.244.0.6 dpu-node-mt2402xz0f9n-mt2402xz0f9n <none> <none>
dpf-operator-system dpu-cplane-tenant1-doca-hbn-jxkxw-ds-ftnpn 2/2 Running 0 2m54s 10.244.1.5 dpu-node-mt2402xz0f8g-mt2402xz0f8g <none> <none>
dpf-operator-system dpu-cplane-tenant1-doca-hbn-jxkxw-ds-gjsqq 2/2 Running 0 3m21s 10.244.3.4 dpu-node-mt2402xz0f7x-mt2402xz0f7x <none> <none>
dpf-operator-system dpu-cplane-tenant1-doca-hbn-jxkxw-ds-k78vb 2/2 Running 0 2m54s 10.244.2.4 dpu-node-mt2402xz0f80-mt2402xz0f80 <none> <none>
dpf-operator-system dpu-cplane-tenant1-nvidia-k8s-ipam-controller-5c77854fcc-grchr 1/1 Running 0 127m 10.244.0.3 dpu-node-mt2402xz0f9n-mt2402xz0f9n <none> <none>
dpf-operator-system dpu-cplane-tenant1-nvidia-k8s-ipam-node-ds-krgzw 1/1 Running 0 6m53s 10.244.1.2 dpu-node-mt2402xz0f8g-mt2402xz0f8g <none> <none>
dpf-operator-system dpu-cplane-tenant1-nvidia-k8s-ipam-node-ds-pr85m 1/1 Running 0 5m57s 10.244.3.3 dpu-node-mt2402xz0f7x-mt2402xz0f7x <none> <none>
dpf-operator-system dpu-cplane-tenant1-nvidia-k8s-ipam-node-ds-x4lfs 1/1 Running 0 7m1s 10.244.0.2 dpu-node-mt2402xz0f9n-mt2402xz0f9n <none> <none>
dpf-operator-system dpu-cplane-tenant1-nvidia-k8s-ipam-node-ds-zlzvf 1/1 Running 0 6m50s 10.244.2.2 dpu-node-mt2402xz0f80-mt2402xz0f80 <none> <none>
dpf-operator-system dpu-cplane-tenant1-ovs-cni-arm64-bpljq 1/1 Running 0 7m1s 10.0.110.213 dpu-node-mt2402xz0f9n-mt2402xz0f9n <none> <none>
dpf-operator-system dpu-cplane-tenant1-ovs-cni-arm64-gls6h 1/1 Running 0 6m50s 10.0.110.212 dpu-node-mt2402xz0f80-mt2402xz0f80 <none> <none>
dpf-operator-system dpu-cplane-tenant1-ovs-cni-arm64-j8wr4 1/1 Running 0 5m57s 10.0.110.211 dpu-node-mt2402xz0f7x-mt2402xz0f7x <none> <none>
dpf-operator-system dpu-cplane-tenant1-ovs-cni-arm64-kbrrn 1/1 Running 0 6m53s 10.0.110.214 dpu-node-mt2402xz0f8g-mt2402xz0f8g <none> <none>
dpf-operator-system dpu-cplane-tenant1-sfc-controller-node-ds-vmfq4 1/1 Running 0 5m57s 10.0.110.211 dpu-node-mt2402xz0f7x-mt2402xz0f7x <none> <none>
dpf-operator-system dpu-cplane-tenant1-sfc-controller-node-ds-x45nl 1/1 Running 0 6m53s 10.0.110.214 dpu-node-mt2402xz0f8g-mt2402xz0f8g <none> <none>
dpf-operator-system dpu-cplane-tenant1-sfc-controller-node-ds-xskh9 1/1 Running 0 7m1s 10.0.110.213 dpu-node-mt2402xz0f9n-mt2402xz0f9n <none> <none>
dpf-operator-system dpu-cplane-tenant1-sfc-controller-node-ds-zfmt5 1/1 Running 1 (5m46s ago) 6m50s 10.0.110.212 dpu-node-mt2402xz0f80-mt2402xz0f80 <none> <none>
dpf-operator-system kube-flannel-ds-2shh7 1/1 Running 0 7m2s 10.0.110.213 dpu-node-mt2402xz0f9n-mt2402xz0f9n <none> <none>
dpf-operator-system kube-flannel-ds-42mlq 1/1 Running 0 6m54s 10.0.110.214 dpu-node-mt2402xz0f8g-mt2402xz0f8g <none> <none>
dpf-operator-system kube-flannel-ds-m7xgt 1/1 Running 0 5m58s 10.0.110.211 dpu-node-mt2402xz0f7x-mt2402xz0f7x <none> <none>
dpf-operator-system kube-flannel-ds-vd574 1/1 Running 0 6m52s 10.0.110.212 dpu-node-mt2402xz0f80-mt2402xz0f80 <none> <none>
dpf-operator-system kube-multus-ds-d5kb4 1/1 Running 0 6m53s 10.0.110.214 dpu-node-mt2402xz0f8g-mt2402xz0f8g <none> <none>
dpf-operator-system kube-multus-ds-gnv88 1/1 Running 0 6m50s 10.0.110.212 dpu-node-mt2402xz0f80-mt2402xz0f80 <none> <none>
dpf-operator-system kube-multus-ds-l66tm 1/1 Running 0 7m1s 10.0.110.213 dpu-node-mt2402xz0f9n-mt2402xz0f9n <none> <none>
dpf-operator-system kube-multus-ds-mh4cj 1/1 Running 0 5m57s 10.0.110.211 dpu-node-mt2402xz0f7x-mt2402xz0f7x <none> <none>
dpf-operator-system kube-sriov-device-plugin-64c29 1/1 Running 0 7m1s 10.0.110.213 dpu-node-mt2402xz0f9n-mt2402xz0f9n <none> <none>
dpf-operator-system kube-sriov-device-plugin-6js9j 1/1 Running 0 6m50s 10.0.110.212 dpu-node-mt2402xz0f80-mt2402xz0f80 <none> <none>
dpf-operator-system kube-sriov-device-plugin-g5gkx 1/1 Running 0 6m53s 10.0.110.214 dpu-node-mt2402xz0f8g-mt2402xz0f8g <none> <none>
dpf-operator-system kube-sriov-device-plugin-lk4z7 1/1 Running 0 5m57s 10.0.110.211 dpu-node-mt2402xz0f7x-mt2402xz0f7x <none> <none>
kube-system coredns-66bc5c9577-gqn8d 1/1 Running 0 127m 10.244.0.4 dpu-node-mt2402xz0f9n-mt2402xz0f9n <none> <none>
kube-system coredns-66bc5c9577-p2xnm 1/1 Running 0 127m 10.244.1.3 dpu-node-mt2402xz0f8g-mt2402xz0f8g <none> <none>
kube-system kube-proxy-64865 1/1 Running 0 5m58s 10.0.110.211 dpu-node-mt2402xz0f7x-mt2402xz0f7x <none> <none>
kube-system kube-proxy-hvjjp 1/1 Running 0 6m52s 10.0.110.212 dpu-node-mt2402xz0f80-mt2402xz0f80 <none> <none>
kube-system kube-proxy-qfbwh 1/1 Running 0 6m54s 10.0.110.214 dpu-node-mt2402xz0f8g-mt2402xz0f8g <none> <none>
kube-system
kube-proxy-w9gg4 1/1 Running 0 7m2s 10.0.110.213 dpu-node-mt2402xz0f9n-mt2402xz0f9n <none> <none>
**Congratulations! The DPF system with the HBN service has been successfully installed.**
### Zero-Trust Mode Checking
Here's a step-by-step procedure to check the Zero-Trust Mode on your NVIDIA BlueField DPU from the host server, including the installation of the Mellanox Firmware Tools (MFT).
> **Warning:** Ubuntu 24.04 was installed on the servers.
1. **Navigate to the NVIDIA 下载 Site:** Open your web browser and go to the official [NVIDIA Mellanox software downloads page](https://network.nvidia.com/products/adapter-software/firmware-tools/).
2. Select the Latest Version for your OS:
3. Transfer and Extract MFT Tools on the Worker 1 BareMetal Host.
**First Pod Console**
```bash
root@worker1:~# tar -xvzf /tmp/mft-4.33.0-169-x86_64-deb.tgz
-
Navigate into the Extracted Directory.
First Pod Console
root@worker1:~# cd mft-4.33.0-169-x86_64-deb/ -
Run following commands.
First Pod Console
root@worker1:~# apt-get install gcc make dkms root@worker1:~# ./install.sh -
Start MST (Mellanox Software Tools) Service and Identify DPU Device Name.
First Pod Console
root@worker1:~# mst start Starting MST (Mellanox Software Tools) driver set Loading MST PCI module - Success Loading MST PCI configuration module - Success Create devices Unloading MST PCI module (unused) - Success root@worker1:~# mst status MST modules: ------------ MST PCI module is not loaded MST PCI configuration module loaded MST devices: ------------ /dev/mst/mt41692_pciconf0 - PCI configuration cycles access. domain:bus:dev.fn=0000:2b:00.0 addr.reg=88 data.reg=92 cr_bar.gw_offset=-1 Chip revision is: 01 -
Perform Zero-Trust Checking.
First Pod Console
root@worker1:~# mlxprivhost -d 2b:00.0 q Host configurations ------------------- level : RESTRICTED Port functions status: ----------------------- disable_rshim : TRUE disable_tracer : TRUE disable_port_owner : TRUE disable_counter_rd : TRUE #Expected Zero-Trust Output.This is the most definitive confirmation.
level : RESTRICTEDmeans the host is in Zero-Trust Mode, and theTRUEflags confirm individual security restrictions are active. -
Verify the host cannot reset DPU firmware:
root@worker1:~# sudo mlxfwreset -d 2b:00.0 -y -l 3 resetExpected output on a Zero-Trust host: -E- Failed to send Register MFRL: Register access Method not supported (264).
The MFRL (Master Firmware Reset Lock) register is access-gated by RESTRICTED mode. Failure here confirms the host cannot trigger a DPU firmware reset.
-
Check Firmware Access with
mlxfwmanager:First Pod Console
root@worker1:~# mlxfwmanager -d 2b:00.0 --query Querying Mellanox devices firmware ... Device #1: ---------- Device Type: BlueField3 Part Number: -- Description: PSID: PCI Device Name: 2b:00.0 Base MAC: N/A Versions: Current Available FW -- Status: Failed to open deviceThe behaviour of
mlxfwmanager --querydepends on the MFT version installed:- MFT < 4.33 returns
Status: Failed to open device— the host cannot read inventory at all in Zero-Trust mode. - MFT 4.33+ returns inventory data (FW/PXE/UEFI versions, PSID, MAC) with
Status: No matching image found. The BMC fulfils the read from cached inventory; this is not a Zero-Trust failure. TheAvailable: N/Aand theNo matching image foundstatus both confirm the host has no path to upload firmware. Write-side operations (--update,mlxfwreset) remain blocked.
Either output is consistent with Zero-Trust mode. Use the
mlxprivhost -d <dev> pwrite probe in step 11 (below) for the definitive verification. - MFT < 4.33 returns
-
Check Device Configuration with
mlxconfig:First Pod Console
root@worker1:~# mlxconfig -d 2b:00.0 q Device #1: ---------- Device type: BlueField3 Name: 900-9D3B6-00CV-A_Ax Description: NVIDIA BlueField-3 B3220 P-Series FHHL DPU; 200GbE (default mode) / NDR200 IB; Dual-port QSFP112; PCIe Gen5.0 x16 with x16 PCIe extension option; 16 Arm cores; 32GB on-board DDR; integrated BMC; Crypto Enabled Device: 2b:00.0 Configurations: Next Boot ... ALLOW_RD_COUNTERS True(1) # No RO, but restricted by mlxprivhost ... PORT_OWNER True(1) # No RO, but restricted by mlxprivhost ... TRACER_ENABLE True(1) # No RO, but restricted by mlxprivhostMost configuration parameters are prefixed with
RO(Read-Only) — the host literally cannot change them, by design. A small number of parameters related to host-side control (PORT_OWNER,ALLOW_RD_COUNTERS,TRACER_ENABLE) are not markedRO, andmlxconfig setwill appear to succeed against them on a Zero-Trust host:root@worker1:~# sudo mlxconfig -d 2b:00.0 set TRACER_ENABLE=0...Apply new Configuration? (y/n) [n] : yApplying... Done!-I- Please power cycle machine to load new configurations.However, the change is functionally a no-op. At runtime, the
mlxprivhostRESTRICTED layer (visible in step 7'sdisable_tracer: TRUE) overrides whatevermlxconfigsays. On the next DPU re-provision, the DPF operator'sDPUFlavor.spec.nvconfigre-applies the canonical settings anyway. The fact thatmlxconfig setreturns "Applying... Done!" does not indicate Zero-Trust is off. -
Step 11. Verify the host cannot escape Zero-Trust mode (write probe with
mlxprivhost):root@worker1:~# sudo mlxprivhost -d 2b:00.0 pExpected output on a Zero-Trust host: -E- Operation is not permitted (refer to the DPU user manual)
The
pargument asks the host to switch its privilege level back to PRIVILEGED. On a Zero-Trust DPU this must fail — that's the whole point of the RESTRICTED level. If this command succeeds and the level flips, the DPU is not in Zero-Trust mode and the DPUFlavor (spec.dpuMode) needs to be re-checked.This is the only command in this section that is fundamentally write-side and therefore unaffected by MFT-version changes to read-side behaviour.
-
Check Low-Level Hardware Access with
ethtool:First Pod Console
root@worker1:~# ethtool -d ens1f0np0 Cannot get register dump: Operation not supportedThis confirms the DPU is preventing deep, low-level hardware access from the host, aligning with Zero-Trust's isolation goals.
Conclusion
The host is operating in Zero-Trust Mode when ALL of the following are true:
| # | Test | Expected on Zero-Trust host |
|---|---|---|
| 1 | mlxprivhost q -> level line |
RESTRICTED |
| 2 | mlxprivhost q -> disable_* flags |
TRUE for all |
disable_rshim, disable_tracer, disable_port_owner, disable_counter_rd all TRUE
| Step | Command | Expected Output |
|---|---|---|
| 3 | mlxconfig q -> count of RO rows |
majority of parameters prefixed RO |
| 4 | ethtool -d <iface> |
Operation not supported |
| 5 | mlxprivhost p (write probe — definitive) |
Operation is not permitted |
| 6 | mlxfwreset reset (optional write check) |
Method not supported |
Tests 1–4 are read-side checks and confirm the configuration is in place. Test 5 (mlxprivhost p) is the authoritative proof: a Zero-Trust host cannot remove its own restrictions. Test 6 reinforces this for firmware-reset specifically.
This means the host has significantly restricted privileges and cannot perform sensitive operations on the DPU, ensuring its security and isolation.
Infrastructure Bandwidth & Latency Validation
Verify the deployment and confirm that the DPU system achieves link-speed performance and low latency by running various tests:
- Iperf TCP—for bandwidth measurements
- RDMA—for bandwidth and latency measurements
- Network isolation
Each test is described in detail. At the end of each test, the achieved performance is displayed.
Notes Make sure that the servers are tuned for maximum performance (not covered in this document).
Performance and Isolation Tests
Now that the test deployment is running, perform bandwidth and latency performance tests between two bare-metal workload servers.
Ubuntu 24.04 was installed on the servers.
-
Before running the tests, check the Gateway address on each HBN pod:
Jump Node Console
$ ki -n dpf-operator-system get pod -o wide | grep doca-hbn dpu-cplane-tenant1-doca-hbn-jxkxw-ds-5sbzs 2/2 Running 0 15m 10.244.0.6 dpu-node-mt2402xz0f9n-mt2402xz0f9n <none> <none> dpu-cplane-tenant1-doca-hbn-jxkxw-ds-ftnpn 2/2 Running 0 15m 10.244.1.5 dpu-node-mt2402xz0f8g-mt2402xz0f8g <none> <none> dpu-cplane-tenant1-doca-hbn-jxkxw-ds-gjsqq 2/2 Running 0 16m 10.244.3.4 dpu-node-mt2402xz0f7x-mt2402xz0f7x <none> <none> dpu-cplane-tenant1-doca-hbn-jxkxw-ds-k78vb 2/2 Running 0 15m 10.244.2.4 dpu-node-mt2402xz0f80-mt2402xz0f80 <none> <none> $ ki exec -it -n dpf-operator-system dpu-cplane-tenant1-doca-hbn-jxkxw-ds-gjsqq -- bash Defaulted container "doca-hbn" out of: doca-hbn, hbn-sidecar, hbn-init (init) root@dpu-cplane-tenant1-doca-hbn-jxkxw-ds-gjsqq:/tmp# ip a s ... 9: vlan11@br_default: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 9216 qdisc noqueue master RED state UP group default qlen 1000 link/ether 0a:ff:4e:3e:99:24 brd ff:ff:ff:ff:ff:ff inet 10.0.121.2/29 scope global vlan11 valid_lft forever preferred_lft forever inet6 fe80::8ff:4eff:fe3e:9924/64 scope link valid_lft forever preferred_lft forever ... $ exit $ ki exec -it -n dpf-operator-system dpu-cplane-tenant1-doca-hbn-jxkxw-ds-k78vb -- bash Defaulted container "doca-hbn" out of: doca-hbn, hbn-sidecar, hbn-init (init) root@dpu-cplane-tenant1-doca-hbn-jxkxw-ds-k78vb:/tmp# ip a s ... 9: vlan11@br_default: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 9216 qdisc noqueue master RED state UP group default qlen 1000 link/ether 0e:7d:99:41:2e:11 brd ff:ff:ff:ff:ff:ff inet 10.0.121.10/29 scope global vlan11 valid_lft forever preferred_lft forever inet6 fe80::c7d:99ff:fe41:2e11/64 scope link valid_lft forever preferred_lft forever ... $ exit -
Connect to a first Workload Server console, install iperf, perftest, check DPU High Speed Interfaces, set route to ethernet and identify the relevant RDMA device:
First Pod Console
root@worker1:~# apt install iperf3 root@worker1:~# apt install perftest root@worker1:~# ip a s ... 6: ens1f0np0: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN group default qlen 1000 link/ether 58:a2:e1:73:69:e6 brd ff:ff:ff:ff:ff:ff altname enp43s0f0np0 ... root@worker1:~# ip route add 10.0.123.0/22 via 10.0.121.2 depuser@worker2:~$ ping 8.8.8.8 PING 8.8.8.8 (8.8.8.8) 56(84) bytes of data. 64 bytes from 8.8.8.8: icmp_seq=1 ttl=117 time=5.35 ms 64 bytes from 8.8.8.8: icmp_seq=2 ttl=117 time=5.10 ms 64 bytes from 8.8.8.8: icmp_seq=3 ttl=117 time=5.15 ms root@worker1:~# rdma link | grep ens1f0np0 link mlx5_0/1 state DOWN physical_state DISABLED netdev ens1f0np0 -
Configure the
ens1f0np0interface on Ubuntu 24.04 usingiproute2. Configuration Overview:Interface IP Address Default Gateway ens1f0np0 10.0.121.1/29 10.0.121.2/29 First Pod Console
# Bring up physical interfaces root@worker1:~# ip link set dev ens1f0np0 up # Assign IP addresses root@worker1:~# ip addr add 10.0.121.1/29 dev ens1f0np0 # Set default route root@worker1:~# ip route add default via 10.0.121.2 dev ens1f0np0 -
Using another console window, reconnect to the jump node and connect to a second Workload Server. From within the servers, install iperf, perftest, check DPU High Speed Interfaces, set route to ethernet and identify the relevant RDMA device:
First Pod Console
root@worker2:~# apt install iperf3 root@worker2:~# apt install perftest root@worker2:~# ip a s ... 6: ens1f0np0: <BROADCAST,MULTICAST> mtu 9000 qdisc noop state DOWN group default qlen 1000 link/ether 58:a2:e1:73:6a:58 brd ff:ff:ff:ff:ff:ff altname enp43s0f0np0 ... root@worker2:~# ip route add 10.0.123.0/22 via 10.0.121.10 depuser@worker2:~$ ping 8.8.8.8 PING 8.8.8.8 (8.8.8.8) 56(84) bytes of data. 64 bytes from 8.8.8.8: icmp_seq=1 ttl=117 time=5.35 ms 64 bytes from 8.8.8.8: icmp_seq=2 ttl=117 time=5.10 ms 64 bytes from 8.8.8.8: icmp_seq=3 ttl=117 time=5.15 ms root@worker2:~# rdma link | grep ens1f0np0 link mlx5_0/1 state DOWN physical_state DISABLED netdev ens1f0np0 -
Configure the
ens1f0np0interface on Ubuntu 24.04 usingiproute2.Configuration Overview
Interface IP Address Default Gateway ens1f0np0 10.0.121.9/29 10.0.121.10/29 Second Pod Console
# Bring up physical interfaces root@worker2:~# ip link set dev ens1f0np0 up # Assign IP addresses root@worker2:~# ip addr add 10.0.121.9/29 dev ens1f0np0 # Set default route root@worker2:~# ip route add default via 10.0.121.10 dev ens1f0np0 -
Run iperf3 TCP bandwidth test between the two workload servers:
On Second Pod (server):
root@worker2:~# iperf3 -sOn First Pod (client):
root@worker1:~# iperf3 -c 10.0.121.9 -t 30 -P 4Expected output (example):
[ ID] Interval Transfer Bitrate Retr [ 5] 0.00-30.00 sec 35.0 GBytes 10.0 Gbits/sec 0 sender [ 5] 0.00-30.00 sec 35.0 GBytes 10.0 Gbits/sec receiver -
Run RDMA bandwidth test using
ib_write_bw:On Second Pod (server):
root@worker2:~# ib_write_bw -d mlx5_0 --report_gbits -FOn First Pod (client):
root@worker1:~# ib_write_bw -d mlx5_0 --report_gbits -F 10.0.121.9Expected output (example):
#bytes #iterations BW peak[Gb/sec] BW average[Gb/sec] MsgRate[Mpps] 65536 5000 98.5 98.2 0.187 -
Run RDMA latency test using
ib_write_lat:On Second Pod (server):
root@worker2:~# ib_write_lat -d mlx5_0 -FOn First Pod (client):
root@worker1:~# ib_write_lat -d mlx5_0 -F 10.0.121.9Expected output (example):
#bytes #iterations t_min[usec] t_max[usec] t_typical[usec] 2 1000 1.02 1.15 1.05 -
Verify network isolation by ensuring that traffic from one tenant cannot reach another tenant's network:
- From a workload server in tenant A, attempt to ping a workload server in tenant B.
- The ping should fail (no response).
Example:
root@worker1:~# ping 10.0.122.1 # IP of a server in a different tenant PING 10.0.122.1 (10.0.122.1) 56(84) bytes of data. ^C --- 10.0.122.1 ping statistics --- 5 packets transmitted, 0 received, 100% packet loss, time 4096msThis confirms that the DPF Zero Trust configuration enforces network isolation between tenants.
| Interface | IP Address | Default Gateway |
|---|---|---|
| ens1f0np0 | 10.0.121.9/29 | 10.0.121.10/29 |
First Pod Console
# Bring up physical interfaces
root@worker2:~# ip link set dev ens1f0np0 up
# Assign IP addresses
root@worker2:~# ip addr add 10.0.121.9/29 dev ens1f0np0
# Set default route
root@worker2:~# ip route add default via 10.0.121.10 dev ens1f0np0
iPerf TCP Bandwidth Test
Move back to the first server console.
-
Start the
iperf3server side:First BM Server Console
root@worker1:~# iperf3 -s ------------------------------------------------------------ Server listening on TCP port 5001 TCP window size: 128 KByte (default) ------------------------------------------------------------ -
Move to the second server console. Start the
iperfclient side:Second BM Server Console
root@worker2:~# iperf3 -c 10.0.121.1 -P 16 ------------------------------------------------------------ Client connecting to 10.0.121.1, TCP port 5001 TCP window size: 16.0 KByte (default) ------------------------------------------------------------ [ 9] local 10.0.121.9 port 48620 connected with 10.0.121.1 port 5001 (icwnd/mss/irtt=14/1448/827) [ 10] local 10.0.121.9 port 48610 connected with 10.0.121.1 port 5001 (icwnd/mss/irtt=14/1448/881) [ 1] local 10.0.121.9 port 48712 connected with 10.0.121.1 port 5001 (icwnd/mss/irtt=14/1448/608) [ 14] local 10.0.121.9 port 48728 connected with 10.0.121.1 port 5001 (icwnd/mss/irtt=14/1448/722) [ 11] local 10.0.121.9 port 48710 connected with 10.0.121.1 port 5001 (icwnd/mss/irtt=14/1448/870) [ 4] local 10.0.121.9 port 48622 connected with 10.0.121.1 port 5001 (icwnd/mss/irtt=14/1448/945) [ 7] local 10.0.121.9 port 48690 connected with 10.0.121.1 port 5001 (icwnd/mss/irtt=14/1448/906) [ 15] local 10.0.121.9 port 48736 connected with 10.0.121.1 port 5001 (icwnd/mss/irtt=14/1448/689) [ 2] local 10.0.121.9 port 48616 connected with 10.0.121.1 port 5001 (icwnd/mss/irtt=14/1448/796) [ 3] local 10.0.121.9 port 48618 connected with 10.0.121.1 port 5001 (icwnd/mss/irtt=14/1448/940) [ 12] local 10.0.121.9 port 48706 connected with 10.0.121.1 port 5001 (icwnd/mss/irtt=14/1448/892) [ 16] local 10.0.121.9 port 48696 connected with 10.0.121.1 port 5001 (icwnd/mss/irtt=14/1448/810) [ 8] local 10.0.121.9 port 48626 connected with 10.0.121.1 port 5001 (icwnd/mss/irtt=14/1448/801) [ 6] local 10.0.121.9 port 48692 connected with 10.0.121.1 port 5001 (icwnd/mss/irtt=14/1448/891) [ 5] local 10.0.121.9 port 48624 connected with 10.0.121.1 port 5001 (icwnd/mss/irtt=14/1448/931) [ 13] local 10.0.121.9 port 48686 connected with 10.0.121.1 port 5001 (icwnd/mss/irtt=14/1448/903) [ ID] Interval Transfer Bandwidth [ 3] 0.0000-10.0058 sec 14.1 GBytes 12.1 Gbits/sec [ 13] 0.0000-10.0057 sec 14.2 GBytes 12.2 Gbits/sec [ 7] 0.0000-10.0056 sec 13.4 GBytes 11.5 Gbits/sec [ 12] 0.0000-10.0057 sec 15.2 GBytes 13.1 Gbits/sec [ 4] 0.0000-10.0058 sec 14.1 GBytes 12.1 Gbits/sec [ 11] 0.0000-10.0058 sec 15.8 GBytes 13.6 Gbits/sec [ 8] 0.0000-10.0057 sec 13.9 GBytes 11.9 Gbits/sec [ 9] 0.0000-10.0058 sec 13.8 GBytes 11.9 Gbits/sec [ 15] 0.0000-10.0057 sec 14.3 GBytes 12.3 Gbits/sec [ 16] 0.0000-10.0058 sec 14.6 GBytes 12.5 Gbits/sec [ 1] 0.0000-10.0057 sec 14.6 GBytes 12.6 Gbits/sec [ 6] 0.0000-10.0058 sec 13.1 GBytes 11.3 Gbits/sec [ 14] 0.0000-10.0059 sec 13.6 GBytes 11.6 Gbits/sec [ 10] 0.0000-10.0055 sec 13.5 GBytes 11.6 Gbits/sec [ 2] 0.0000-10.0057 sec 14.0 GBytes 12.0 Gbits/sec [ 5] 0.0000-10.0058 sec 14.6 GBytes 12.6 Gbits/sec [SUM] 0.0000-10.0010 sec 227 GBytes 195 Gbits/sec
RoCE Latency Test
Return to the first server console.
-
Start the
ib_read_latserver side:First BM Server Console
root@worker1:~# ib_read_lat -F -n 20000 -d mlx5_0 ************************************ * Waiting for client to connect... * ************************************ -
Move to the second server console. Start the
ib_read_latclient side:Second BM Server Console
root@worker2:~# ib_read_lat -F -n 20000 -d mlx5_0 10.0.121.1 --------------------------------------------------------------------------------------- RDMA_Read Latency Test Dual-port : OFF Device : mlx5_0 Number of qps : 1 Transport type : IB Connection type : RC Using SRQ : OFF PCIe relax order: ON ibv_wr* API : ON TX depth : 1 Mtu : 1024[B] Link type : Ethernet GID index : 3 Outstand reads : 16 rdma_cm QPs : OFF Data ex. method : Ethernet --------------------------------------------------------------------------------------- local address: LID 0000 QPN 0x0048 PSN 0x77ae88 OUT 0x10 RKey 0x186ded VAddr 0x005fe0b3e3a000 GID: 00:00:00:00:00:00:00:00:00:00:255:255:10:00:121:09 remote address: LID 0000 QPN 0x0048 PSN 0x51948d OUT 0x10 RKey 0x186ded VAddr 0x00577584a67000 GID: 00:00:00:00:00:00:00:00:00:00:255:255:10:00:121:01 --------------------------------------------------------------------------------------- #bytes #iterations t_min[usec] t_max[usec] t_typical[usec] t_avg[usec] t_stdev[usec] 99% percentile[usec] 99.9% percentile[usec] 2 20000 3.98 65.30 4.08 7.89 7.17 31.51 36.33 ---------------------------------------------------------------------------------------
RoCE Bandwidth Test
Return to the first server console.
-
Start the
ib_write_bwserver side:First BM Server Console
root@worker1:~# ib_write_bw -s 1048576 -F -D 30 -q 64 -d mlx5_0 ************************************ * Waiting for client to connect... * ************************************ -
Move to the second server console. Start the
ib_write_bwclient side:Second BM Server Console
root@worker2:~# ib_write_bw -s 1048576 -F -D 30 -q 64 -d mlx5_0 10.0.121.1 --report_gbit --------------------------------------------------------------------------------------- RDMA_Write BW Test Dual-port : OFF Device : mlx5_0 Number of qps : 64 Transport type : IB Connection type : RC Using SRQ : OFF PCIe relax order: ON ibv_wr* API : ON TX depth : 128 CQ Moderation : 1 Mtu : 1024[B] Link type : Ethernet GID index : 3 Max inline data : 0[B] rdma_cm QPs : OFF Data ex. method : Ethernet --------------------------------------------------------------------------------------- … --------------------------------------------------------------------------------------- #bytes #iterations BW peak[Gb/sec] BW average[Gb/sec] MsgRate[Mpps] 1048576 420000 0.00 220.72 0.026312 ---------------------------------------------------------------------------------------
Network Isolation Test
Finally, verify that the two servers running on different networks—using virtual functions on PF0 and PF0 can't communicate with each other.
Connect to the first workload server, with the PF0 network, and try to ping the PF0 on second node.
- Run the
pingcommands from
RDG for DPF Zero Trust (DPF-ZT) with HBN DPU Service
PF0 to PF0:
First BM Server Console
root@worker1:~# ping -c 3 10.0.121.9
PING 10.0.121.9 (10.0.121.9) 56(84) bytes of data.
64 bytes from 10.0.121.9: icmp_seq=1 ttl=62 time=0.896 ms
64 bytes from 10.0.121.9: icmp_seq=2 ttl=62 time=0.241 ms
64 bytes from 10.0.121.9: icmp_seq=3 ttl=62 time=0.258 ms
- Try to ping the PF0 on nodes 3 and 4. Run the
pingcommands from PF0 to PF0:
First BM Server Console
root@worker1:~# ping -c 3 10.0.122.1
PING 10.0.122.1 (10.0.122.1) 56(84) bytes of data.
From 10.0.121.2 icmp_seq=1 Destination Host Unreachable
From 10.0.121.2 icmp_seq=2 Destination Host Unreachable
From 10.0.121.2 icmp_seq=3 Destination Host Unreachable
--- 10.0.122.1 ping statistics ---
3 packets transmitted, 0 received, +3 errors, 100% packet loss, time 2045ms
root@worker1:~# ping -c 3 10.0.122.9
PING 10.0.122.9 (10.0.122.9) 56(84) bytes of data.
From 10.0.121.2 icmp_seq=1 Destination Host Unreachable
From 10.0.121.2 icmp_seq=2 Destination Host Unreachable
From 10.0.121.2 icmp_seq=3 Destination Host Unreachable
--- 10.0.122.9 ping statistics ---
3 packets transmitted, 0 received, +3 errors, 100% packet loss, time 2067ms
This ping operation should fail due to the network isolation implemented in HBN using different VLANs, VNIs and VRFs.
Done.
Authors
 |
Boris KovalevBoris Kovalev has worked for the past several years as a 解决方案 Architect, focusing on NVIDIA Networking/Mellanox technology, and is responsible for complex machine learning, Big Data and advanced VMware-based cloud research and design. Boris previously spent more than 20 years as a senior consultant and solutions architect at multiple companies, most recently at VMware. He has written multiple reference designs covering VMware, machine learning, Kubernetes, and container solutions which are available at the NVIDIA Documents website. |
NVIDIA, the NVIDIA logo, and BlueField are trademarks and/or registered trademarks of NVIDIA Corporation in the U.S. and other countries. Other company and product names may be trademarks of the respective companies with which they are associated.™
2025 NVIDIA Corporation. All rights reserved.©