News

Call for Papers

Download CFP as TXT file

Important Dates

Papers due: 2 Feb 2017 (AoE)
13 Jan 2017
Notification: 24 Feb 2017
Workshop Date: 2 Jun 2017

Links

IPDPS 2017

Previous Workshops

REPPAR 2016
REPPAR 2015
REPPAR 2014

Previous Proceedings

Euro-Par Workshops (incl. REPPAR) 2015
Euro-Par Workshops (incl. REPPAR) 2014

Workshop Program

Workshop Date: 2 June 2017
Place: TBA

Time Talk
08:45-09:00 Workshop Introduction and Opening
Session chair: Sascha Hunold
09:00-10:00 Keynote: Spack and software reproducibility in HPC
Todd Gamblin (LLNL)
Session chair: Sascha Hunold
Abstract: see here
10:00-10:30 Coffee break
10:30-12:00 Session 1
Session chair: Arnaud Legrand
  1. The Popper Convention: Making Reproducible Systems Evaluation Practical Ivo Jimenez, Michael Sevilla, Noah Watkins, Carlos Maltzahn, Jay Lofstead, Kathryn Mohror, Andrea Arpaci-Dusseau and Remzi Arpaci-Dusseau.
  2. Towards Trustworthy Testbeds thanks to Throughout Testing Lucas Nussbaum.
  3. Examining the Reproducibility of Using Dynamic Loop Scheduling Techniques in Scientific Applications Franziska Hoffeins, Florina M. Ciorba and Ioana Banicescu.
12:00-14:00 Lunch break
14:00-15:00 Session 2
Session chair: Lucas Nussbaum
  1. Characterizing the Performance of Modern Architectures Through Opaque Benchmarks: Pitfalls Learned the Hard Way Luka Stanisic, Lucas Mello Schnorr, Augustin Degomme, Franz C. Heinrich, Arnaud Legrand and Brice Videau.
  2. Towards Reproducible Blocked LU Factorization Roman Iakymchuk, Enrique S. Quintana-Orti, Erwin Laure and Stef Graillat.
15:00-15:30 Discussion and Workshop Closing
Session chair: Sascha Hunold
15:30-16:00 Coffee break

Spack and software reproducibility in HPC

Todd Gamblin (LLNL)

Abstract

Reproducibility is particularly important for scientific research, as it enables scientists to verify and build on past work. Unfortunately, scientific software is notoriously difficult to build, even though reproducing the software environment is fundamental for reproducing a prior experiment. Package management tools can help by codifying the builds of common software. Indeed, reproducible builds have recently been a focal point for many Linux distributions, driven by verifiability and security. However, reproducibility requirements for mainstream package managers are different than those for scientific computing. Hardware in HPC environments is more diverse than on typical Linux machines, and bitwise reproducibility is not always necessary or desirable. Spack is a package manager for HPC that aims to provide reproducibility across a diverse range of platforms and architectures. This talk will discuss the larger context of reproducible builds and how Spack fits into it. The talk will focus on the guarantees Spack provides, how Spack’s build process is isolated (or not) from the host, and why reproducing and validating software on HPC platforms is hard.