Exascale Computing Project (ECP) Software Technologies (ST) software, Extreme-scale Scientific Software Stack (E4S) v21.11, includes a subset of ECP ST software products, and demonstrates the target approach for future delivery of the full ECP ST software stack. Also available are a number of ECP ST software products that support a Spack package, but are not yet fully interoperable. As the primary purpose of the v21.11 is demonstrating the ST software stack release approach, not all ECP ST software products were targeted for this release. Software products were targeted primarily based on existing Spack package maturity, location within the scientific software stack, and ECP SDK developer experience with the software. Each release will include additional software products, with the ultimate goal of including all ECP ST software products.
The current E4S container offerings include Docker images based on Red Hat Enterprise Linux 7, Red Hat Enterprise Linux 8, Ubuntu 18.04 (Bionic), and Ubuntu 20.04 (Focal Fossa) for Continuous Integration. Our images can run on X86_64, PPC64LE, and AARCH64 depending on the particular image. Our full E4S Release images (not for Continuous Integration) are based on Ubuntu 18.04 (ppc64le) and Ubuntu 20.04 (x86_64). In addition to offering a full E4S image containing a comprehensive selection of E4S software released on a quarterly cycle, we also offer a set of minimal base images suitable for use in Continuous Integration (CI) pipelines where Spack is used to build packages.
Docker images are available on the E4S Docker Hub.
Recipes which can be used to build some of our images from scratch are available on the E4S GitHub repository.
Our recipes make use of Spack packages available as pre-built binaries in the E4S Build Cache.
Spack contains packages for all of the products listed in the E4S 21.11 Full Release category (see above Release Notes). General instructions for building software with Spack can be found at the Spack website. For more information, see /usr/local/packages/ecp in the container referenced here. Questions concerning building those packages are deferred to the associated package development team.
Container images contain binary versions of the Full Release packages listed above. The full-featured GPU-enabled container image is available from Dockerhub:
# docker pull ecpe4s/e4s-gpu
This is a full-featured multi-arch image.
It contains all relevant GPU stacks, the full set of E4S release packages installed w/ Spack, Python machine learning packages optimized for GPU, and TAU.
These images come with MPICH, CMake, and the relevant GPU SDK -- either ROCM, NVIDIA CUDA Toolkit and NVHPC, or Intel OneAPI -- available as LMOD modules
These are barebones operating system images which contain only essential build tools and python packages needed by Spack.
These images are intended to be used in continuous integration workflows where Spack is first cloned and then used to build and test software.
The E4S 21.11 release is also available on AWS as an EC2 AMI with ID ami-0db9d49091db1c25f in the US-West-2 (Oregon) region.