All Episodes

September 2, 2020 52 min
In episode 30, we interviewed Todd Gamblin from the Lawrence Livermore National Laboratory about the Spack project. We discussed his current research project along with his involvement in Spack. We widely discussed the philosophy of Spack, some usage patterns, its capabilities for managing package management in HPC clusters as well as standalone computers and which operating systems it supports at the moment. Todd shared with us his opinion on the trend for containerized workloads to achieve reproducible science and why it may not be the goal we need to set. He highlighted for us the similarities and differences between EasyBuild and Spack as well as the origin of those differences. We finished the interview with our usual quick questions. 00:00:00 Intro music 00:00:17 Introduction 00:00:36 Introducing Todd Gamblin 00:00:58 His current research topics 00:01:23 Spack as official duties 00:01:43 Spack usage at Lawrence Livermore National Laboratory 00:02:01 Other research projects 00:02:47 Profiling in HPC 00:04:24 His role as leader of software packaging technology for the exascale computing project 00:04:58 One-minute elevator pitch for Spack 00:05:34 Spack's usage philosophy compared to other package managers 00:06:59 Installation from source code or binary? 00:07:28 Spack's usage in the top500 super computers 00:07:49 Geographical distribution of users 00:08:18 Number of packages in the repo and some examples 00:09:05 Managing computer clusters with Spack's automation capabilities 00:11:04 Module files in Spack 00:12:32 Syntax of a Spack package file 00:13:43 Configuration of compiler flags 00:15:00 Importing python libraries in the Spack files 00:15:48 The procedure to submit a package 00:16:27 Review process for new packages 00:17:34 Reasons for rejection of Spack packages 00:18:01 Operating systems supported by Spack 00:18:23 WSL and Spack 00:18:58 Restricting packages to certain hardware and software configurations 00:20:04 Build testing and nightly builds 00:21:28 Working with containers in a Spack environment 00:22:25 Deploying prebuilt containers 00:23:05 About the "universality" of containers 00:24:16 His opinion on containerized applications for reproducible science 00:26:17 Spack's log file to document reproducibility 00:27:13 Reproducing older results 00:28:10 Specifying requirements on compilers 00:30:39 Post-installation verification test 00:31:10 Using Spack on a standalone computer instead of HPC systems 00:32:56 Differences between EasyBuild and Spack 00:34:24 EasyBuild in the top500 00:34:49 Transitionning between EasyBuild and Spack 00:35:38 Other alternatives 00:36:23 Using EasyBuild and Spack on the same system 00:38:36 When did the project start? 00:39:53 External contributions to Spack 00:40:53 How many core developpers? 00:41:30 Organization of the community and governance model 00:43:06 Who decides which package is accepted in the repo? 00:44:38 Spack's choice of software license 00:47:09 Todd's vision about the importance of FLOSS for the openness of science 00:48:13 Possible negative impacts of FLOSS 00:48:58 Most notable recent scientific discovery 00:49:14 Favourite text processing tool 00:49:25 A topic in science about which he recently changed his mind about 00:49:58 Anything else we forgot to ask? 00:50:09 How to contact Todd 00:50:46 Conclusion
Share
Mark as Played

Advertise With Us

Popular Podcasts

Crime Junkie
The Daily

The Daily

This is what the news should sound like. The biggest stories of our time, told by the best journalists in the world. Hosted by Michael Barbaro. Twenty minutes a day, five days a week, ready by 6 a.m.

Dateline NBC

Dateline NBC

Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations.

For You

    Music, radio and podcasts, all free. Listen online or download the iHeart App.

    Connect

    © 2021 iHeartMedia, Inc.

    • Help
    • Privacy Policy
    • Terms of Use
    • AdChoicesAd Choices