Spack Community BOF
Tuesday, June 10, 2025 1:00 PM to 2:00 PM · 1 hr. (Europe/Berlin)
Hall G1 - 2nd floor
Birds of a Feather
Community EngagementCompiler and Tools for Parallel Programming
Information
Spack is a package manager for scientific computing. With nearly 1,500 contributors from academia, industry, and government laboratories, Spack is a broad collaboration with a wide range of use cases, from small-scale development on laptops and clusters, to software release management, to software deployment on many of the top supercomputer sites in the world.
At ISC 2025, we will announce the release of Spack v1.0, which will feature some of the most extensive changes to Spack since its inception 12 years ago. This is the first 1.0 version after years at version 0.x. Spack v1.0 will feature:
- Compilers and their runtimes modeled as dependencies
- A stable package API, and a separate package repository from the core Spack tool
- A faster, parallel build scheduler and on-node dynamic build job scheduling
Modeling compilers in Spack has been a long and difficult road, and we are excited to finally release compiler dependencies to users. This change will reduce the amount of configuration required to bootstrap Spack, enable users to more easily and safely mix compilers, and greatly simplify binary package installation and portability. We will give an overview of how these features will benefit users, as well as an overview of how to migrate from old Spack versions to the new one. We will also have updates on our stable package API and the guarantees we are making around package stability and reproducibility over time. Finally, we will briefly demonstrate on-node multi-package parallel builds and the speedups we have seen using them.
Following the highlights, we will shift to questions and an interactive user survey. All are invited to provide feedback, request features or changes, and participate in the discussion! Users will be encouraged to discuss specific use-cases and we will have Spack developers on-stage ready to answer and questions from the community.
Join the BOF and help us make HPC software installation simple!
Organizers:
At ISC 2025, we will announce the release of Spack v1.0, which will feature some of the most extensive changes to Spack since its inception 12 years ago. This is the first 1.0 version after years at version 0.x. Spack v1.0 will feature:
- Compilers and their runtimes modeled as dependencies
- A stable package API, and a separate package repository from the core Spack tool
- A faster, parallel build scheduler and on-node dynamic build job scheduling
Modeling compilers in Spack has been a long and difficult road, and we are excited to finally release compiler dependencies to users. This change will reduce the amount of configuration required to bootstrap Spack, enable users to more easily and safely mix compilers, and greatly simplify binary package installation and portability. We will give an overview of how these features will benefit users, as well as an overview of how to migrate from old Spack versions to the new one. We will also have updates on our stable package API and the guarantees we are making around package stability and reproducibility over time. Finally, we will briefly demonstrate on-node multi-package parallel builds and the speedups we have seen using them.
Following the highlights, we will shift to questions and an interactive user survey. All are invited to provide feedback, request features or changes, and participate in the discussion! Users will be encouraged to discuss specific use-cases and we will have Spack developers on-stage ready to answer and questions from the community.
Join the BOF and help us make HPC software installation simple!
Organizers:
Format
On Site
Targeted Audience
Spack has a broad audience, including HPC users who simply install and run packages, developers who author their own packages and automate their dependency management, HPC facility staff who deploy large software stacks with packages and modules, and team deployment specialists who deploy software stacks targeted towards a specific team.
Speakers
Todd Gamblin
Computer ScientistLawrence Livermore National Laboratory
Gregory Becker
Computer ScientistLawrence Livermore National Laboratory
Massimiliano Culpo
Computer Scientistn.p. complete, S.r.l.
Adam Stewart
Post-docTechnical University of Munich
Kathleen Shea
Software DeveloperLawrence Livermore National Laboratory