Conference paper Open Access

Investigating Trade-offs between Portability, Performance and Maintainability in Exascale Systems

Elvira-Maria Arvanitou; Apostolos Ampatzoglou; Nikolaos Nikolaidis; Aggeliki-Agathi Tzintzira; Areti Ampatzoglou; Alexander Chatzigeorgiou

Due to the rapid advancements in the hardware architectures of High-Performance Computing infrastructures, new challenges have arisen in the development of scientific software applications. In particular, software that runs on Exascale machines, needs to be highly portable, highly parallelizable and at the same time maintainable, since software for HPC evolves constantly over time. By taking into account that an overall optimization of all the aforementioned qualities is not realistic, in this study, we explore the possible trade-offs, when optimizing the run-time qualities of the software (i.e., performance and portability) through state-of-practice techniques in Exascale software development, in expense of code maintainability, as expressed by technical debt. To achieve this goal, we have performed a case study, in which the effect of run-time optimizations on technical debt has been measured. The results suggest that run-time optimizations tend to reduce TD principal, whereas the effect on interest is not consistent. The results are discussed in detail in this paper from the point of view of both researchers and practitioners.

Files (316.6 kB)
Name Size
publication-5.pdf
md5:636e5497c21f010f40c8cc346cbe1a0e
316.6 kB Download
27
28
views
downloads
Views 27
Downloads 28
Data volume 8.9 MB
Unique views 25
Unique downloads 27

Share

Cite as