Quickstart Guide
The Quickstart guide is intended for deployment on dedicated nodes or virtual machines (VMs) running an RPM-based Linux distribution on x86_64 architecture.
A minimum high-availability (HA) master node configuration requires:
Three stateful nodes (two permanent and one temporary; the temporary node may also be made permanent)
At least one stateless compute node
Optional stateless GPU node
Hardware Requirements
Both master nodes must include a 25 GB `/drbd` partition to support data replication and synchronization for high availability. In addition, ensure the head node has at least 60 GB of free disk space in the directory where the OpenCHAI repository and RPM stack are cloned. This space is required for offline installation.
Minimum hardware requirements for testing:
Head Node - 4 CPU cores - 4 GB RAM - (Recommended: 8 CPU cores and 8 GB RAM)
Master Nodes - 8–12 CPU cores - 8 GB RAM (each)
Software Requirements
View the OpenCHAI Software Stack
OpenCHAI Manager Tool Setup
This section provides a step-by-step guide for installing the Open CDAC HPC-AI Manager Tool (OpenCHAI) on a bare-metal server.
The guide focuses on a quick installation workflow with minimal explanation. A moderately experienced cluster administrator can use these steps to deploy and configure a standard HPC-AI cluster without referring to the full OpenCHAI Administrator Manual.
Installing the Head Node
Step 1. Install Git
Install the Git package on the head node:
yum install git
Step 2. Clone the OpenCHAI Repository
Clone the repository into a directory with at least 60 GB of free disk space. This space is required to store the offline software stack RPMs.
git clone https://github.com/OpenHPC-AI/OpenCHAI.git
# Navigate to the OpenCHAI directory
cd OpenCHAI
Step 3. Configure the OpenCHAI Manager Tool
3.1 Ensure that the OpenCHAI software stack tar file is available in the directory below to enable a smooth and reliable setup.
Offline Mode (Recommended if Available)
Copy the OpenCHAI software stack for Alma Linux or Rocky Linux from removable media (USB drive, local disk, or external storage) into:
ls -la ./hpcsuite_registry/hostmachine_reg/<os-version>/
For manual package downloads, refer to the HPC-Sangrah Vault:
HPC-Sangrah Vault
Online Mode
If the stack is not available locally, the OpenCHAI Manager Tool can download the required packages from the network during Step 3.3 of the setup process.
3.2 Ansible Inventory Setup
Configure the Ansible inventory on the head node to enable communication with all service nodes in the HPC-AI cluster, including:
HPC master nodes
Management nodes
AI master nodes
Login nodes
BMC nodes
Edit the inventory file according to your environment:
vim chai_setup/inventory_def.txt
3.3 Proceed with CHAI Manager Head Node Setup
Once the inventory is configured, initialize the CHAI Manager Head Node. This step prepares the primary control node for managing cluster services and orchestration workflows.
During the OpenCHAI Manager head-node setup, you are presented with an option to select the OpenCHAI software packages to be installed. Review the available package options carefully and choose the configuration that best aligns with your target environment, deployment model, and workload requirements.
bash ./configure_openchai_manager.sh
If OpenCHAI packages are already available locally, the setup proceeds faster. Otherwise, the installer provides an option to download them during execution.
Step 4. Inventory Verification
Verify inventory configuration:
ansible-inventory --list
Step 5. Update Cluster Environment Variables
Update all HPC-AI cluster environment variables:
bash ./chai_setup/update_group_var_all.sh
Note
Ensure public network access is enabled on all service nodes so that missing packages can be installed from public repositories.
Step 6: Configure the PXE Server for Service Node Booting
Follow this section to set up the PXE server, which enables network booting and automated installation of service nodes.
Define DHCP Clients
Create or edit the DHCP client input file:
vim ./servicenodes/dhcp_clients_mac.txt
Format:
<node_name>,<mac_address>,<ip_address>
Example:
master01,00:0C:29:43:18:B4,172.10.3.101
master02,00:0C:29:43:18:B5,172.10.3.102
login01,00:0C:29:43:18:B6,172.10.3.103
This file is used to generate static DHCP reservations.
PXE server setup
Run the PXE server setup script:
bash servicenodes/pxe_server_setup.sh headnode
During execution
All DHCP clients are displayed in a table
You are prompted to update MAC or IP values if required
Final confirmation is requested before execution
For detailed PXE server configuration, refer to the link below.
Step 7. HPC-AI Head Node Setup
Run the head node setup script:
bash ./servicenodes/headnode_node_setup.sh
Step 8. Inventory Verification
Verify connectivity with all cluster hosts:
ansible all -m ping
# Verify connectivity using a specific SSH port
ansible all -m ping -e "ansible_port=22"
If communication issues occur, update the inventory file and reapply:
bash ./chai_setup/update_inventory_def.sh
—
Step 9. HPC Master Nodes Deployment and Configuration
Note
Currently, the platform supports both online and offline deployment modes in a mixed configuration. Ensure that all service nodes (including both master nodes) have public network connectivity available. Additionally, verify that time synchronization (NTP/Chrony) is properly configured and consistent across all nodes.
To deploy and configure HPC master nodes, execute:
bash ./servicenodes/hpc_master_ha_node_setup.sh hpc_master
Feedback
Was this page helpful?