Cloud HPC Toolkit uses an HPC blueprint
to define a deployment. For each blueprint, you can use the
setting to define a set of modules that can be used to customize your
deployment. These modules are used to specify options such as your compute
resources, networking, schedulers, monitoring applications, and file systems.
This document outlines the core supported module options for Cloud HPC Toolkit v1.3.0 or later. For a complete list of supported modules including the experimental options, see the modules page on the Cloud HPC Toolkit GitHub repository documentation.
Supported file systems
The following file systems are supported:
You can also specify that Cloud HPC Toolkit integrates a pre-existing file system into your deployment.
The following schedulers are supported:
Supported Compute Engine resources
The following features are supported for Compute Engine resources:
VM creation - many of the core VM customization options are supported including, but not limited to, the following:
- Machine type options: all machine types
- VM instance placement policy options: both compact and spread placement policies
- GPU integration: all GPU types
- Advanced networking: all options including gVNIC support and Tier 1 higher bandwidths
- Service account setup
- Disabling simultaneous multithreading (SMT)
- Startup scripts
- Spot VMs
Supported monitoring options
The following tools are supported for collecting measurements of your service and of the Google Cloud resources that you use.
Supported networking resources
The following features are supported for Google Cloud's VPC resources:
Supported software installation and system setup
Cloud HPC Toolkit can be used for the following use cases:
- Automation of application installations by using Spack
- Ansible installation
- Google Cloud's operations suite Ops Agent installation and setup
- Network File System (NFS) client installation and automatic mounting
- Custom image building automation with Packer
- Review how to prepare your HPC blueprint.
- Try a quickstart tutorial, see Deploy an HPC cluster with Slurm.