Abstract
Recent work has introduced a number of tools and techniques for reasoning about the interplay between application performance and portability, or “performance portability”. These tools have proven useful for setting goals and guiding high level discussions, but our understanding of the performance
portability problem remains incomplete. Different views of the same performance efficiency data offer different insights into an application’s performance portability (or lack thereof): standard statistical measures such as the mean and standard deviation require careful interpretation, and even metrics designed specifically to measure performance portability may obscure differences between applications. This paper offers a critical assessment of existing approaches for summarizing performance efficiency data across different platforms, and proposes visualization as a means to extract useful information about the underlying distribution. We explore a number of alternative visualizations, outlining a new methodology that enables developers to reason about the performance portability of their applications and how it might be improved. This study unpicks what it might mean to be “performance portable” and
provides useful tools to explore that question.
portability problem remains incomplete. Different views of the same performance efficiency data offer different insights into an application’s performance portability (or lack thereof): standard statistical measures such as the mean and standard deviation require careful interpretation, and even metrics designed specifically to measure performance portability may obscure differences between applications. This paper offers a critical assessment of existing approaches for summarizing performance efficiency data across different platforms, and proposes visualization as a means to extract useful information about the underlying distribution. We explore a number of alternative visualizations, outlining a new methodology that enables developers to reason about the performance portability of their applications and how it might be improved. This study unpicks what it might mean to be “performance portable” and
provides useful tools to explore that question.
Original language | English |
---|---|
Title of host publication | 2020 IEEE/ACM International Workshop on Performance, Portability and Productivity in HPC (P3HPC 2020) |
Subtitle of host publication | Supercomputer 2020 |
Publisher | Institute of Electrical and Electronics Engineers (IEEE) |
ISBN (Print) | 9780738110899 |
Publication status | Published - 13 Nov 2020 |
Event | Performance, Portability, and Productivity in HPC Forum 2020 - Duration: 1 Sep 2020 → 2 Sep 2020 https://p3hpcforum2020.alcf.anl.gov/ |
Conference
Conference | Performance, Portability, and Productivity in HPC Forum 2020 |
---|---|
Abbreviated title | P3HPC 2020 |
Period | 1/09/20 → 2/09/20 |
Internet address |
Keywords
- high performance computing
- performance portability