Dynamic Multi-Objective Workflow Scheduling for Cloud Computing Based on Evolutionary Algorithms

Abstract

Cloud computing is a dominant heterogeneous and distributed system, that offers on-demand resource capacity for different requirements of customers. Cloud workflow scheduling is a largely studied research area that targets efficient utilization of cloud resources. In this paper, we model a dynamic workflow scheduling problem, which is among the first attempts that incorporate dynamism on resource failures and changing number of objectives. A set of dynamic multi-objective evolutionary algorithms from the literature are utilized for dynamic workflow scheduling problem, where four of them are variants of the commonly used NSGA-II algorithm (DNSGA-II-HM, DNSGA-II-A, DNSGA-II-B, DNSGA-II-RI) and the remaining one is dynamic extension of the multi-objective particle swarm optimization algorithm (DMOPSO). In our experimental study, five different objectives are considered, which are minimization of the makespan, the cost and the energy, and maximization of the reliability and the resource utilization. The empirical study of the given five algorithms is conducted with real-world applications from Pegasus workflow management systems, where the DNSGA-II-B procedure outperforms the other alternatives for most of the test instances, based on the number of non-dominated solutions, the Schott's spacing and the hyper-volume metrics.