Getting Started with AWS ParallelCluster

AWS ParallelCluster is a powerful tool for deploying and managing High Performance Computing (HPC) clusters in the AWS cloud. This guide walks through the essential steps to create your first HPC cluster using ParallelCluster.

Prerequisites

Before you begin, ensure you have:

  • AWS CLI installed and configured with appropriate permissions
  • Basic understanding of Linux and SSH
  • Python 3.11 or newer

Step 1: ParallelCluster Installation

Let’s start by creating a dedicated Python virtual environment for ParallelCluster:

python3.11 -m venv ~/Envs/ParallelCluster-01
source ~/Envs/ParallelCluster-01/bin/activate

Next, install ParallelCluster in the virtual environment (we’ll use version 3.12 for this example):

(ParallelCluster-01)$ pip install aws-parallelcluster==3.12

Note: If you encounter issues with setuptools, you may need to fix the version:

pip install setuptools==69.5.1

Verify your installation:

(ParallelCluster-01)$ pcluster version
{
  "version": "3.12.0"
}

Step 2: Verify AWS CLI Configuration

Before proceeding, ensure your AWS CLI is properly configured:

$ aws sts get-caller-identity

You should see output similar to one of these examples:

For SSO login (using Identity Management):

{
    "UserId": "AAAAAAAAAAAAAAAAAAAAA:jouser",
    "Account": "111111111111",
    "Arn": "arn:aws:sts::111111111111:assumed-role/AWSnnnnSSO_admin_1111111111111111/jouser"
}

For local credentials with IAM role:

111122223333    arn:aws:iam::111122223333:user/jouser           XXXXXXXXXXXXXXXXXXXXX

Step 3: Create the ParallelCluster VPC

ParallelCluster can automatically create a VPC with appropriate networking for your cluster. This is the recommended approach for most users:

(ParallelCluster-01)$ pcluster configure --config cluster-01.yaml

When prompted, select the following options:

  • Automate VPC creation: y
  • Choose an Availability Zone (e.g., us-west-2a)
  • Network Configuration: Head node in a public subnet and compute fleet in a private subnet

This creates a secure network architecture where:

  • The head node is accessible via SSH from the internet (with security group restrictions)
  • Compute nodes are in a private subnet with no direct internet access
  • NAT Gateway enables compute nodes to access the internet for updates and package installation

After completion, you’ll have a configuration file (cluster-01.yaml) with subnet IDs for both the head node and compute fleet:

HeadNode:
  Networking:
    SubnetId: subnet-11111111111111111

Scheduling:
  Scheduler: slurm
  SlurmQueues:
  - Name: queue01
    Networking:
      SubnetIds:
        - subnet-22222222222222222

Step 4: Customize Your Cluster Configuration

The auto-generated configuration is a starting point, but you’ll want to customize it for your specific workload. Here’s an example configuration for a medium-sized HPC cluster with dedicated storage and compute nodes:

Region: us-west-2
Image:
  Os: rhel8
HeadNode:
  InstanceType: m6id.2xlarge
  DisableSimultaneousMultithreading: false
  Ssh:
    KeyName: ec2-key-pdx
  Networking:
    ElasticIp: true
    SubnetId: subnet-XXXXXXXXXXXXXXXXX
    AdditionalSecurityGroups:
      - sg-XXXXXXXXXXXXXXXXX
AdditionalPackages:
  IntelSoftware:
    IntelHpcPlatform: false
Scheduling:
  Scheduler: slurm
  SlurmSettings:
    ScaledownIdletime: 10
    Dns:
      DisableManagedDns: true
  SlurmQueues:
  - Name: storage01
    CapacityType: ONDEMAND
    Networking:
      SubnetIds:
        - subnet-XXXXXXXXXXXXXXXXX
      PlacementGroup:
        Enabled: true
        Name: cluster-01-placement-group-01
    ComputeResources:
    - Name: storage
      InstanceType: m6idn.2xlarge
      MinCount: 8
      MaxCount: 16
      DisableSimultaneousMultithreading: false
  - Name: batch01
    CapacityType: ONDEMAND
    Networking:
      SubnetIds:
        - subnet-XXXXXXXXXXXXXXXXX
      PlacementGroup:
        Enabled: true
        Name: cluster-01-placement-group-01
    ComputeResources:
    - Name: name
      InstanceType: m6idn.xlarge
      MinCount: 16
      MaxCount: 32
      DisableSimultaneousMultithreading: true

Key configuration elements:

  1. Instance Types:
    • Head node: m6id.2xlarge provides a balance of compute, memory, and local storage
    • Storage nodes: m6idn.2xlarge with enhanced network performance
    • Compute nodes: m6idn.xlarge optimized for compute workloads
  2. Security:
    • ElasticIP for consistent access
    • Additional security groups for organizational requirements
    • Private subnet for compute nodes
  3. Performance:
    • Placement groups for low-latency networking
    • Selective disabling of simultaneous multithreading
    • Instance types with local NVMe storage
  4. Scaling:
    • Minimum and maximum node counts for each resource type
    • ScaledownIdletime for cost optimization

Security Warning: By default, the security group created for the head node allows SSH traffic from any source (0.0.0.0/0). This inbound rule should be updated to a specific IP address or range.

Step 5: Create the Cluster

Now that you have a customized configuration, create your cluster:

(ParallelCluster-01)$ pcluster create-cluster -n test-cluster01 -c cluster-01.yaml --rollback-on-failure false

The --rollback-on-failure false flag is recommended for initial deployments as it allows for easier troubleshooting if issues arise.

What’s Next?

After your cluster is created, you’ll have a basic HPC environment with:

  • A head node accessible via SSH
  • Compute nodes managed by the Slurm scheduler
  • Basic job submission capabilities

However, to transform this into a production-ready HPC environment, you’ll need additional customization. Check out my follow-up post on Customizing an HPC cluster with ParallelCluster to learn how to implement advanced cluster customization techniques.

This post is part of a series on building production-ready HPC environments on AWS.