학술논문

Three level benchmarking of Singularity containers for scientific calculations.
Document Type
Article
Source
Annales Mathematicae et Informaticae. 2023, Vol. 58, p133-146. 14p.
Subject
*CONTAINERIZATION
*CONTAINERS
*MEMORY
*COMPUTER software
Language
ISSN
1787-5021
Abstract
In this work we present our results from benchmarking Singularity containers running scientific calculations intended for HPC at three levels of complexity, comparing them against native software environments. Our investigations were run on up-to-date hardware and the latest available software as of early 2023. To get a more detailed picture, we examined three different system aspects: we ran separate benchmarks on CPU, memory, and I/O intensive operations. These aspects were tested with microbenchmarks, and at a high level - an HPC workload pipeline which intensely loaded each aspect. As a result of our investigations we show that Singularity containerization continues to provide as good, or in some cases even better performance indicators as in a native environment - with the substantial added benefits of application flexibility and portability. [ABSTRACT FROM AUTHOR]