7. Personalized Configuration

7.1 Using Network Interface Bonding to Accelerate Deployment

Using network interface bonding to increase the management node's upload bandwidth can improve deployment speed.

1. On the management node, install the ifenslave.

sudo apt install ifenslave

2. load bonding module。

# load bonding module
sudo modprobe bonding
# 
sudo lsmod | grep bonding

3. Configure network interface bonding. For example, configure a bond in mode 6 (balance-alb) using two interfaces named eth0 and eth1. Modifying /etc/netplan/00-installer-config.yaml:

/etc/netplan/00-installer-config.yaml

network:
  version: 2
  ethernets:
    eth0:
      dhcp4: false
    eth1:
      dhcp4: false
  bonds:
    bond0:
      addresses: [192.168.1.88/24]
      gateway4: 192.168.1.1
      nameservers:
        addresses:
          - 114.114.114.114
      interfaces:
        - eth0
        - eth1
      parameters:
        mode: balance-alb

Then run:

sudo netplan apply

4. Check the Bonding Status:

# Check the Bonding Status:
cat /proc/net/bonding/bond0 
sudo ethtool bond0

To view the configuration and status information of the bonded network interface bond0, including speed, duplex mode, link status, and hardware features.

5. Delete Bond

Comment out the added content in the /etc/netplan/00-installer-config.yaml file.

sudo ifconfig bond0 down
sudo rmmod bonding

Check:

ifconfig bond0

7.2 Upgrading CUDA Version

The CUDA version can be upgraded to match your downloaded CUDA runfile. Simply modify the CUDA field in workspace/version.yaml to match the downloaded filename:

CUDA: cuda_13.0.3_580.126.20_linux.run  # default

For example, to upgrade to another CUDA version:

CUDA: cuda_13.2.1_595.58.03_linux.run

WARNING

Ensure the GPU Driver version is compatible with the CUDA version.

Download CUDA runfile:

cd podsys-bxx/workspace/drivers
wget https://developer.download.nvidia.com/compute/cuda/13.2.1/local_installers/cuda_13.2.1_595.58.03_linux.run