Cost effective hybrid genetic algorithm for scheduling scientific workflows in cloud under deadline constraint

Abstract

Cloud has emerged as a convenient platform for executing complicated scientific applications from multiple disciplines by providing on-demand and scalable infrastructure on rental basis. Research and scientific community often opt for workflows to model these scientific applications Workflow scheduling has been extensively studied for decades with regard to grid and cluster computing, but few initiatives have been tailored for cloud. What's more, the previous work fails to incorporate the basic principles of IaaS clouds like pay-as-you-go model, elasticity, heterogeneity, dynamic provisioning and issues of VM's performance variation and acquisition delay besides other QoS requirements. This paper proposes a resource provisioning and scheduling strategy using genetic algorithm with the aim to optimise the overall execution cost while staying below the given deadline. The performance is further enhanced by using a high quality seed generated by predict earliest finish time (PEFT) algorithm which acts as a catalyst and helps the algorithm to converge faster. The proposed approach is simulated in WorkflowSim and evaluated using various well-known different sized realistic scientific workflows. The results validate the better performance of our approach over numerous state-of-art-algorithms.