Document Type

Thesis

Publication Date

4-2018

Disciplines

Computer Sciences | Systems Architecture

Advisor

Michael Heroux, Computer Science

Abstract

Scientific workflows for high-performance computing (HPC) are becoming increasingly complex. Developing a way to simplify these workflows could save many hours for both HPC users and developers, potentially eliminating any time spent managing software dependencies and experiment set-up. To accomplish this, we propose using two programs together: Docker and Spack. Docker is a container platform and Spack is a package manager designed specifically for HPC. In this paper, we show how Docker and Spack can be used to containerize the extreme-scale Scientific Software Development Kit (xSDK). Doing this makes the xSDK far more accessible to non-computer scientists and lowers time spent by developers on dependency management. Implementing a system such as this on a large scale could change the functioning of the HPC industry.

Share

COinS