Gravitational search algorithm based novel workflow scheduling for heterogeneous computing systems

• Published in: Simulation modelling practice and theory Vol. 96; p. 101932
• Main Authors: Biswas, Tarun, Kuila, Pratyay, Ray, Anjan Kumar, Sarkar, Mayukh
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.11.2019
• Subjects: Energy-Consumption; GSA; HCSs; Load-Balancing; Makespan; Workflow-Scheduling; Load-Balancing; Makespan; HCSs; GSA; Workflow-Scheduling; Energy-Consumption
• ISSN: 1569-190X; 1878-1462
• DOI: 10.1016/j.simpat.2019.101932

•A novel GSA based workflow scheduling for HCSs.•A recursive algorithm to generate a valid execution sequence.•Novel agent representation by preserving dependency constraints.•Fitness function using Makespan, load-balancing, and energy-consumption.•Simulation and hypothesis testing through ANOVA. The significance of workflow applications (WAs) in various domains like, scientific experiments, research, education, health-care, etc. is increasing with the technological advancements of modern computing world. The WA consists of a set of tasks with complex dependency relationships. Generating a valid execution sequence by preserving precedence constraints is challenging. Workflow scheduling algorithms (WSAs) are invoking more attention as one of the real time concerns to the researchers. Although, number of research attitudes have been shown for WSA, still it is hard to design one single coherent algorithm that satisfies multiple criteria simultaneously. Moreover, WSA is well known due to its non-deterministic polynomial (NP)-hard nature. In this paper, we have proposed a gravitational search algorithm (GSA) based workflow scheduling for heterogeneous computing systems. The proposed work considers multiple conflicting objectives which are minimization of makespan, load-balancing, and energy-consumption. A novel representation of agents is demonstrated by preserving dependency constraints amongst the tasks. A recursive algorithm is designed to generate a valid execution sequence of tasks that helps to restrict the precedence relationship. Derivation of fitness function is done based on the considered multiple objectives. The performances are analyzed and validated by extensive simulations on different set of scientific, Fast Fourier Transformation (FFT), and synthetic workflow applications data set. It is observed that the proposed GSA shows considerable improvements in terms of the considered objectives over recent GSA based approaches like, gravitation search algorithm for load scheduling (GSAL) and hybrid gravitational search algorithm (HGSA). The results are also validated using a statistical hypothesis test, Analysis of Variance (ANOVA) to demonstrate the effectiveness of the proposed work.