Posted On: Sep 14, 2020
Amazon ParallelCluster is a fully supported and maintained open source cluster management tool that makes it easy for scientists, researchers, and IT administrators to deploy and manage High Performance Computing (HPC) clusters in the Amazon Web Services cloud. HPC clusters are collections of tightly coupled compute, storage, and networking resources that enable customers to run large scale scientific and engineering workloads.
Significant feature enhancements to this latest release of ParallelCluster include:
- Support for multiple instance types in Slurm: Users can now create multiple job submission queues inside of ParallelCluster and specify multiple instance types within each queue. On job submission, users can specify the instance types they would like to use for their job as well as the queue to which they would like to submit it, simplifying multi-stage workflows and the ability to run multiple workloads with distinct needs from a single cluster.
- Support for Elastic File System: Elastic File System (EFS) is now officially supported inside of ParallelCluster for use as a network file system with a shared directory between the cluster’s master and compute nodes.
- Slurm power management plugin: We have rearchitected our integration with Slurm to use the Slurm power management plugin. This integration improves cluster stability and precision in scaling compute resources to match individual jobs’ requirements.
Amazon ParallelCluster is available at no additional charge, and you pay only for the Amazon Web Services resources needed to run your applications. Amazon ParallelCluster is released via the Python Package Index (PyPI). Amazon ParallelCluster’s source code is hosted under the Amazon Web Services repository on GitHub at here, where you can learn how to launch your own HPC cluster on Amazon Web Services.
For more detail you can find the complete release notes for the latest version of Amazon ParallelCluster here.