The goal of this research work is to reduce wait time of HPC (high performance computing) applications in schedulers queue by applying a co-scheduling strategy. This strategy allows the execution of more than one task with different non-overlapping requirements for computational resources simultaneously. Co-scheduling strategy reduces task queue wait time and improves utilization of cluster resources when compared to the scheduling strategies that do not allow for parallel task execution on the same machine. We have proposed a method for measuring application processing speed in its run-time, which can be used as a feedback for scheduling strategies. In this work, we have formalized the co-scheduling problem and proposed strategies for solving it. For some strategies we have shown analytically the upper bounds values of their competitive ratios. Besides that for the proposed scheduling strategies we ran numerical experiments using imitation models to show how they compare to the optimal strategy.

Язык оригиналаанглийский
Страницы (с-по)133-137
Число страниц5
ЖурналCEUR Workshop Proceedings
СостояниеОпубликовано - 1 дек 2021
Событие9th International Conference "Distributed Computing and Grid-Technologies in Science and Education", GRID 2021 - Dubna, Российская Федерация
Продолжительность: 5 июл 20219 июл 2021
Номер конференции: 9

Предметные области Scopus

  • Компьютерные науки (все)


Подробные сведения о темах исследования «HPC WORKLOAD BALANCING ALGORITHM for CO-SCHEDULING ENVIRONMENTS». Вместе они формируют уникальный семантический отпечаток (fingerprint).