|
ИСТИНА |
Войти в систему Регистрация |
ИСТИНА ЦЭМИ РАН |
||
Evaluating the Impact of MPI Network Sharing on HPC Applicationsстатья
- Авторы: Khudoleeva Anna, Stefanov Konstantin, Voevodin Vadim
- Сборник: Parallel Computational Technologies. PCT 2023
- Серия: Communications in Computer and Information Science
- Том: 1868
- Год издания: 2023
- Издательство: Springer Nature
- Местоположение издательства: Switzerland
- Первая страница: 3
- Последняя страница: 18
- DOI: 10.1007/978-3-031-38864-4_1
- Аннотация: In any modern supercomputer system, a so-called noise inevitably occurs. It can be defined as an external influence of the software and hardware environment leading to a change in the execution time or other properties of applications running on a supercomputer. Although the noise can noticeably affect the performance of HPC applications in some cases, neither the nature of its occurrence nor the degree of its influence have been investigated in detail. In this paper, we study how much a certain type of noise, caused by sharing of MPI network resources, can impact the performance of parallel programs. To do this, we conducted a series of experiments using synthetic noise on the Lomonosov-2 supercomputer to determine to what extent such noise can slow down the execution of widely used benchmarks and computing cores.
- Добавил в систему: Воеводин Вадим Владимирович