Optimal task scheduling for partially heterogeneous systems

Orr, Michael; Sinnen, Oliver

doi:10.1016/j.parco.2021.102815

Cited by 8 publications

(2 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Multiprocessor systems [2,3] are used to solve wide-ranging complex problems. The availability of multiple processors helps in the parallel execution of tasks onto different processors, which enhances the overall efficiency of the system.…”

Section: Introductionmentioning

confidence: 99%

A Genetics Based Makespan Optimizer with Knowledge Augmented Fuzzy Probabilities

Gupta

Iyer²,

Agarwal

et al. 2023

Preprint

View full text Add to dashboard Cite

Scheduling of various tasks in the multiprocessor environment has been a source of difficult problems for researchers in the field of computer engineering. The widespread issue of multiprocessor task scheduling can be explained as scheduling tasks onto processors and ordering their execution so that several performance standards can be improved. The nondeterministic polynomial-time (NP) completeness of the task scheduling problem has motivated researchers to suggest a variety of heuristic algorithms that can achieve suboptimal results in a reduced amount of polynomial time. Recently, genetic algorithms (GA) have received a lot of responsiveness as GAs are robust and give assurance of a better-quality solution. The outcomes of the experiments carried out with scheduling problems illustrate that genetic algorithms are superior to heuristic algorithms concerning the quality of their outcomes, but they are noticeably slower than heuristic algorithms concerning running time. To overcome the issues related to heuristic algorithms and genetic algorithms, a new approach is proposed in this research work, which combines a GA with various types of scheduling heuristics. In GA, the crossover and mutation rates are the control parameters that need to be set carefully to get optimal results. Generally, these key factors are left to the GA users, but in this research work, a GA that adopts a fuzzy knowledge-based system to control GA key factors has been proposed. To analyze the efficiency of the proposed method, various tests have been performed by varying the number of generations and by analyzing the variations in average schedule length. Based on the findings, the suggested hybrid algorithm is capable of competing with conventional scheduling algorithms in terms of increased quality of outcomes.

show abstract

Section: Introductionmentioning

confidence: 99%

A Genetics Based Makespan Optimizer with Knowledge Augmented Fuzzy Probabilities

Gupta

Iyer²,

Agarwal

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…The heterogeneity coefficient helps KAIROS weight resources differently, which aligns with previous task scheduling algorithms for heterogeneous processors [147,148]. To determine C j , we first set the coefficient of the base instance type (e.g., lowest latency instance) to 1 as a normalization point, then calculate C j as the latency ratio.…”

Section: Kairos Query Distribution Mechanismmentioning

confidence: 93%

Making machine learning on HPC systems cost-effective and carbon-friendly

View full text Add to dashboard Cite

Dissertation xv1.1 This dissertation sets out to optimize machine learning (ML) systems for both costefficiency and environmental sustainability, achieving these goals autonomously without the need for developer or user intervention. It introduces a range of innovations that, when synergized with strategies for cost-informed configuration and carbon-aware scheduling, enable the system to become cost-effective and carbonfriendly, respectively.

show abstract