Introduction to containers
Containers provide a way to package up applications, libraries, scripts, configurations and data in a way that enables applications or even complex workflows to run consistently on a laptop as well as on a supercomputer. A container encapsulates an entire runtime environment that can be transferred, shared and deployed rapidly while being lighter on host system resources than a full virtual machine environment.
Multiple container systems exist, oriented to:
On MeluXina the Singularity Community Edition (SingularityCE) container system is provided to support containerized workloads.
Highlighted features of SingularityCE:
- Easy execution of parallel (MPI) containerized applications
- Support for GPU accelerated applications
- Support for creating and running encrypted containers
- Support for trusted containers that are PGP-signed and verified
- Support for using Docker containers and downloading them from public or private registries
- Support for Open Containers Initiative (OCI) containers
For a complete list of features, check out the SingularityCE documentation.
Users of the Singularity container system can develop their own as well as run pre-existing containers.
The build stage requires administrative (root/
sudo) privileges, and thus has to be performed on a system under user's control (laptop, workstation, server ...).
Once a container has been created, or an existing one selected from a container registry (such as Docker Hub, the Nvidia Cloud (NGC), or a private one), it can be run on one of the MeluXina computational modules.
Examples of using Singularity on MeluXina are provided in the dedicated section.