Created on Jun 30, 2019
BlueField's Ethernet DPU solutions combine Arm® processing power with advanced network offloads to accelerate a multitude of security, networking and storage applications.
Based on the BlueField system-on-chip (SoC) programmable network engine, BlueField DPUs integrate NVIDIA ConnectX network controller with Arm multicore processors and cryptology options to deliver an innovative and high-performance programmable platform.
Through this QSG (Quick Start Guide) article you will be able to quickly set up a simple test lab which will allow you to experiment and explore BlueField DPU capabilities. The setup includes a bare metal server hosting BlueField 25GbE DPU connected back-to-back to another bare metal server with ConnectX-5 NIC.
|2||x86 servers running CentOS 7.5||-|
|1||ConnectX-5 NIC (FW: 16.24.0328 and up)||-|
|1||BlueField 25GbE DPU with Arm processor (FW: 18.24.1000 and up)||-|
|1||25GB/s SFP28 networking cable||-|
|1||QSA28 Adapter (100GB-to-25GB Adapter)||-|
|1||UART-to-USB cable||For DPU USB console access|
|1||Mini USB Type B cable||For DPU USB Console Access|
|1||bfb image to be loaded on the BlueField DPU Arm.|| Image contains:|
Physical Configuration Diagram
The following diagram shows the physical configuration between the BlueField DPU and the ConnectX-5 NIC.
Physical Configuration Instructions
- Insert ConnectX-5 NIC into Server A
- Insert BlueField DPU into Server B
- Connect the DPU UART cable from DPU to its hosting Server B USB port #1
- Connect Mini USB cable from DPU to its hosting Server B USB port #2
- Insert QSA28 Adapter into Server B 100GB port
- Connect one of the DPU ports to the ConnectX-5 using 25GB cable
- Connect both servers to a management LAN
DPU Server Preparations
Install useful packages:
Install MLNX_OFED with the mentioned flags below.
- Use automatic FM update during MLNX_OFED installation ONLY if the current BlueField FW version is 18.24.1000 (GA) or above. In case its older, manually burn the FW after MLNX_OFED installation.
- In case you plan to run DPDK on the host itself, install MLNX_OFED with "--dpdk --upstream-libs" flags
- The "bluefield" flag is not required in some of the MLNX_OFED versions and OS releases
Enable SR-IOV on FW (Optional):
Uncomment the following line in /etc/modprobe.d/rshim.conf to set the default RShim access method to the DPU:
- When this line is commented, the first available driver will be used.
- When the mini USB cable is not used, the driver should be rshim_pcie
Create a file named ifcfg-tmfifo_net0 under /etc/sysconfig/network-scripts/ directory, with the following content:
ifcfg-tmfifo_net0 is the network interface used for IP communication between the x86 host and the DPU with rshim_net driver.
Enable ip forwarding and set a NAT rule to allow the DPU OS to communicate with the outer world via the host management interface:
Disable Firewall and Network Manager services
- Reboot the server. Once server is up again, run some configuration sanity tests:
To verify BlueField FW version:
To verify RShim modules are loaded:
To identify the backend active RShim driver:
To verify the NAT rule for DPU communication with the outer world:
Contact NVIDIA Support to download the BlueField Quick Start bfb image
Place the bfb image on the host and load it on the BlueField DPU using the following command:
Use one of the console access methods described in the "DPU Management Methods" section below to follow the image installation process.
When bfb image installation process is completed, connect to the DPU with the console or SSH, as described below, and test connectivity from the SmarNIC OS to the outer world.
DPU Management Methods
To manage the DPU from the hosting server, use one of the following methods.
Console over USB-Serial / Serial Port properties: Rate 115200, HW Flow Control OFF.
Console over USB (rshim_usb) / Serial Port properties: Rate 115200, HW Flow Control OFF
Basic DPU Management Operations
How-to soft reset the DPU via host server:
Issue the following command on the server:
Follow the boot process using console access:
This is not a power cycle of the DPU.
How-to load BlueField DPU OS image from hosting server:
- Place bfb image on the server
Issue the following command on the server:
Follow the installation and boot process using the console access:
BlueField DPU can operate in two modes:
- SEPERATED_HOST (default mode)
In this mode, both the x86 hosting server and the DPU Arm OS act as separated entities; Both can use and communicate with each other or the network via the ConnectX-5 module of the DPU.
In this mode, the x86 hosting server communicates with the outer world only through the DPU Arm.
In order to identify the DPU current mode, issue the following commands on the x86 hosting server:
Switching to EMBEDDED Mode from SEPERATED_HOST Mode
On the x86 host:
Enable EMBEDDED mode:
On the Arm:
Issue the following commands to enable ECPF parameters (applied per port):
NoteThe ECPF parameters will be available for setting only after completing the previous steps.
- Reboot the x86 host.
- Verify configuration:
Once the procedure is completed, issue the following command on both the x86 host and the Arm:
On x86 host: the output should list two net devices.
On Arm: the output should list four representors:
Issue the following commands on the Arm to see the correlation between PCI device and its representors:
Switching to SEPERATED_HOST Mode from EMBEDDED Mode
On the x86 host:
Enable SEPERATED_HOST mode:
Testing Traffic in EMBEDDED Mode using OVS Offload
In Embedded mode, traffic from the x86 server hosting the DPU to the remote x86 server hosting the ConnectX-5 is going via the DPU Arm. OVS with offload capabilities is used to forward the traffic.
Follow the steps below to test traffic from x86 DPU hosting server, via Arm with OVS offload, to the remote ConnectX-5 server:
On the Arm OS create OVS bridge with the 2 representors of the used port and Offload capabilities:
- Set IP addresses on interface ens1f0 of the x86 host and on the ConnectX-5 host.
- Test ping between x86 host and ConnectX-5 host (it will flow via the Arm OVS). To verify that traffic from Arm OVS is Offloaded to the HW, you can either:
Run tcpdump on the representors - only the first ICMP packets will be captured"
Dump the OVS offloaded flows:
Dump the HW TC flow rules of the representors and look for the "in_hw" flag:
On the ConnectX-5 external host, start iperf server:
On the x86 host start iperf client:
Itai Levy Over the past few years, Itai Levy has worked as a Solutions Architect and member of the NVIDIA Networking “Solutions Labs” team. Itai designs and executes cutting-edge solutions around Cloud Computing, SDN, SDS and Security. His main areas of expertise include NVIDIA BlueField Data Processing Unit (DPU) solutions and accelerated OpenStack/K8s platforms.
Over the past few years, Itai Levy has worked as a Solutions Architect and member of the NVIDIA Networking “Solutions Labs” team. Itai designs and executes cutting-edge solutions around Cloud Computing, SDN, SDS and Security. His main areas of expertise include NVIDIA BlueField Data Processing Unit (DPU) solutions and accelerated OpenStack/K8s platforms.