This repo contains the latest version which is available in the Trovi Artifact titled "MPI and Spack Based HPC Cluster".
The Trovi artifact provides a reproducible, ready-to-use HPC cluster on the Chameleon Cloud, with automated setup of nodes for immediate execution of MPI-based tasks/applications.
Message Passing Interface (MPI) is the backbone of high-performance computing (HPC), enabling efficient scaling across thousands of processing cores.
This project provides a reproducible MPI setup on the Chameleon testbed. It deploys an MPI cluster on a configurable number of nodes. Users login to a "main" node. From there, MPI applications can be built and run across the entire cluster.
- Pre-built Images for Different Workloads: Choose from CPU-only, CUDA, or ROCm images depending on your computation needs. Each image comes pre-configured and is available in the Appliance catalog.
- MPI Ready Cluster: Every node has MPICH, OpenMPI and Spack installed and configured, allowing you to immediately compile and run parallel applications without manual setup.
- Shared Filesystem Support: Chameleon NFS shared filesystem can be set up to easily share data across all nodes in the cluster (available only on sites that support Chameleon Shared File System like CHI@UC).
- Quick Start Examples: Jupyter notebooks using Python-CHI and OpenStack Heat Template for Bare Metal Instances, Openstack Heat Template for KVM are included for rapid experimentation and deployment, helping you get started with minimal effort.