HPC application best practices on AWS

This repository contains HPC application best practices, specifically designed for, and optimized to run on AWS.
In particular, these best practices take into account the peculiarity of AWS HPC-specific services and EC2 instances, in order to get the best out of them.
This repo is maintained by AWS HPC Solution Architects, who will take care of updating and improving these best practices as AWS services evolve or new settings/tunings are discovered. This is not intended to be an AWS supported product or service, though.

HPC application best practices included (to date):

1. ANSYS Fluent
2. Abaqus
3. LS-Dyna
4. Optistruct
5. STAR-CCM+
6. ANSYS CFX
7. ANSYS Mechanical

HPC synthetic benchmarks:

1. STREAM - Memory bandwidth benchmark
2. OSU Micro-Benchmarks - MPI latency, bandwidth, and collective performance benchmarks

HPC tuning and configuration guides:

1. Flexible Cores Configuration - Examples for configuring flexible core counts on Hpc7a and Hpc8a instances using Intel MPI and OpenMPI, including explicit core pinning strategies
2. Test EFA - Scripts and instructions to verify that Elastic Fabric Adapter (EFA) is functioning correctly on your cluster

Request a HPC application best practice

We're starting with the most common HPC applications, specifically in the CAE market.
We're also updating the list of included HPC application best practice regularly, based on your feedback.
So, feel free to request a new HPC application best practice from the ISSUES .
We will do our best to satisfy your requests.

Get Started

On existing Clusters

If you already have a cluster up&running and you want to try these best practices, you can git clone this repository:

git clone https://github.com/aws-samples/hpc-applications.git

Then, (if needed) customize the example launch script for the application you want to try and run it.

Build a new HPC Cluster (1-Click deployment)

You can build a new HPC cluster with just 1-Click using either:

• AWS ParallelCluster - Open source cluster management tool
• AWS Parallel Computing Service (PCS) - Managed HPC cluster service

HPC application Benchmarks

In addition to application best practices, this repo will include some HPC application benchmarks. For all the included applications, we've run some benchmarks using public datasets. We'll publish our data and some charts to show the performance and scalability you should be aiming to achieve.

AWS HPC products and services being used

• AWS ParallelCluster and its Documentation
• AWS Parallel Computing Service (PCS) and its Documentation
• Elastic Fabric Adapter (EFA) and its Documentation
• Amazon FSx for Lustre and its Documentation
• Amazon EFS and its Documentation
• EC2 Image Builder and its Documentation
• AWS EC2 Hpc8a Instances
• AWS EC2 Hpc7a Instances
• AWS EC2 Hpc6a Instances
• AWS EC2 Hpc6id Instances
• AWS EC2 Hpc7g Instances

Security

See CONTRIBUTING for more information.

License

This library is licensed under the MIT-0 License. See the LICENSE file.