Todd Gamblin
Todd Gamblin is a Distinguished Member of Technical Staff in the Livermore Computing division at Lawrence Livermore National Laboratory. He created Spack, a popular open source HPC package management tool with a rapidly growing community of contributors. He leads the Packaging Technologies Project in the U.S. Exascale Computing Project, LLNL's DevRAMP project on developer productivity, and an LLNL Strategic Initiative on software integration. His research interests include dependency management, software engineering, parallel computing, performance measurement, and performance analysis.
Sessions
Dependency solving is at the core of most package managers, and it can be one of the major performance bottlenecks. Keeping a package solver performant as more and more packages, versions, and features are added can be very challenging. This talk is a deep dive into how we’ve done this for Spack.
Spack’s solver has two phases: one that sets up a solve based on metadata from package descriptions, and another to run the solver. We’ll look at performance issues that come from metadata management, including the cost of loading python files and the difficulty of caching metadata from package descriptions written in a pure Python DSL. We’ll also look at the solver itself. Spack’s solver uses Answer Set Programming (ASP), a Prolog-like paradigm for combinatorial problems. We have developed a profiler to look at which parts of our ASP program are exercised by different types of solves, and we’ll discuss a few useful optimizations that this has enabled.
Welcome to the second day of PackagingCon 2023.