A Makespan-optimized Task-Level Scheduling Strategy for Cloud Workflow Systems

Abstract

Cloud-based workflow systems is a platform which can be embedded in a cloud computing infrastructure. Virtual machine has the ability to run multi-tasks simultaneously and time-sharing characteristic, but cannot take advantage of VM’s benefit, therefore, the makespan of scheduling strategy in datacenter cannot be reduced availably. In this paper, we bring up a new makespan model which take advantage of VM’s time-shared characteristic to schedule cloud workflow in task-layer. Furthermore, a novel Ant Colony Optimization (ACO) scheduling strategy is designed to obtain the optimal makespan. This scheduling strategy is implemented in Swinburne Decentralized Workflow for Cloud. The results suggest that by exploiting a time-sharing characteristic of VM, our scheduling strategy offers a significant improvement over the existing approaches including the makespan optimization by scheduling strategy within a datacenter, ACO scheduling strategy can converge fast with different task sets. The makespan is smaller than its counterpart without time-sharing of VMs.