Managing the Scientific Software Ecosystem with Spack

Modern scientific software is far from monolithic -- the largest packages comprise 70 or more dependency libraries. Building and deploying such packages for a variety of architectures, compilers, and dependency libraries is tedious and error-prone, and the complexity of the task impedes reuse and reproducibility. In other fields, package managers are routinely used to manage large collections of software, but HPC software has unique requirements, such as highly specialized compilers and finely tuned, ABI-incompatible interfaces, that make the use of a traditionalpackage manager difficult.Spack is an open source package manager for HPC. Originally developed at Lawrence Livermore National Laboratory, Spack now spans over 30 organizations and over 100 contributors. Spack can install multiple versions, configurations, and finely tuned builds, allowing scientists to rebuild and reuse even the most complex packages across different supercomputers. Spack's dependency model and its support for combinatorial package versioning simplify the use of HPC software for developers, end users, and facilities staff. This poster describes Spack, its adoption in the scientific community this far, and the methods we have used to sustain Spack as a large open source project. p.p1 {margin: 0.0px 0.0px 0.0px 0.0px; font: 12.0px Helvetica; color: #5756d6; -webkit-text-stroke: #5756d6} span.s1 {font-kerning: none} Poster presented at SIAM CSE17 PP108 Minisymposterium: Software Productivity and Sustainability for CSE and Data Science. p.p1 {margin: 0.0px 0.0px 0.0px 0.0px; font: 12.0px Helvetica} p.p2 {margin: 0.0px 0.0px 0.0px 0.0px; font: 12.0px Helvetica; min-height: 14.0px}