How to Mount Google Coral USB on Proxmox

Proxmox VE (Virtual Environment) is a comprehensive, open-source server management platform that combines virtualization technologies like KVM (Kernel-based Virtual Machine) and LXC (Linux Containers). Google Coral USB Accelerator, on the other hand, is a USB-based AI inference accelerator designed to speed up machine learning tasks using Google’s Edge TPU (Tensor Processing Unit). Integrating the two can lead to powerful AI applications in a virtualized environment.

In this article, we’ll guide you through the steps of mounting the Google Coral USB Accelerator to a Proxmox VE environment.

Prerequisites

1. A Proxmox VE server up and running.
2. A Google Coral USB Accelerator.
3. The VM or container in Proxmox where you intend to use the Coral USB Accelerator should have USB passthrough support.

Step-by-Step Guide

1. Connect the Google Coral USB Accelerator:

Connect the Google Coral USB Accelerator to one of the USB ports of your Proxmox VE server.

2. Identify the USB Device:

Before you can pass the device through to a VM, you need to identify it. SSH into your Proxmox server or access its terminal, and use the following command:

lsusb

Look for a device that resembles the Coral USB Accelerator. Note down the device ID which will look something like Bus 001 Device 003.

3. Configure VM for USB Passthrough:

a. Open the Proxmox web interface.

b. Navigate to the specific VM you wish to attach the Coral USB to.

c. Under the ‘Hardware’ tab of the VM, click on the ‘Add’ button and select ‘USB Device’.

d. From the dropdown list, locate and select the Google Coral USB Accelerator. Ensure the ‘USB3’ option is checked if your device and server support USB 3.0 for better performance.

e. Click on the ‘Add’ button.

4. Start the VM:

Start or reboot the VM. Once the VM is up and running, the Coral USB Accelerator should be accessible from within the VM.

5. Install Required Libraries:

Inside the VM, depending on the OS, you’ll need to install the Edge TPU runtime and other required libraries to interact with the Coral device. Typically, for Debian-based distributions, you can follow Google’s instructions, which often look something like:

echo "deb https://packages.cloud.google.com/apt coral-edgetpu-stable main" | sudo tee /etc/apt/sources.list.d/coral-edgetpu.list
curl https://packages.cloud.google.com/apt/doc/apt-key.gpg | sudo apt-key add -
sudo apt-get update
sudo apt-get install libedgetpu1-std

Using Google Coral USB with LXC Containers:

Integrating the Google Coral USB Accelerator with an LXC container in Proxmox is a slightly different process than with VMs, primarily because containers share the host kernel. Here’s a simplified process:

1. Identify the Device: Just as with VMs, first identify the Coral USB device using the lsusb command.
2. Enable USB Passthrough for the Container: Edit the configuration file for the desired container. This is typically located at /etc/pve/lxc/[CTID].conf, where [CTID] is the ID of your container. Add the following line:

lxc.cgroup.devices.allow: c 189:* rwm

This line allows the container to access USB devices.

3. Mount the Device: Add another line to the same configuration file, indicating the Coral USB:

lxc.mount.entry: /dev/bus/usb/[BUS]/[DEVICE] dev/bus/usb/[BUS]/[DEVICE] none bind,optional,create=file

Replace [BUS] and [DEVICE] with the appropriate values you obtained from the lsusb command.

4. Restart the Container: Restart the LXC container to apply these changes.
5. Install Required Libraries: As with VMs, you’ll need to ensure the Edge TPU runtime and other necessary libraries are installed within the container.
6. Test the Device: After setting everything up, it’s a good idea to verify the Coral USB Accelerator’s functionality within the LXC container.

By integrating the Google Coral USB Accelerator into your LXC container, you benefit from the lightweight nature of containers while leveraging AI-enhanced computations. This approach can provide a more efficient way of deploying AI models, especially in clustered or resource-constrained environments.