Look-ahead workflow scheduling with width changing trend in clouds

With the migration of more and more workflows to clouds, cloud workflow scheduling becomes the main bottleneck for meeting user's Quality of Service (QoS) due to the dependency between tasks in a workflow and the elasticity of cloud resources. Besides, the pricing model of cloud resources causes the workflow execution time (WET) and workflow execution cost (WEC) to be critical in workflow scheduling. In this paper, we investigate how to optimize WEC under WET as a deadline constraint for workflow scheduling in clouds and propose a look-ahead workflow scheduling algorithm with width changing trend (W-LA) to solve it. First, we come up with the concept of the width changing trend of a workflow. On the basis of this concept, we define the priority of each task and design a novel deadline distribution strategy to distribute the deadline constraint to each task suitably. Then, we propose a look-ahead instance selection framework (LAISF), where selecting instance not only is based on the impact of the selection on the task being assigned, but also looks ahead in the scheduling to consider the impact of this selection on this task's subsequent tasks. Finally, based on them, W-LA follows a three-step heuristic scheduling: rank tasks by their priorities, distribute the deadline constraint and select instances for tasks by LAISF. W-LA is compared with the state-of-the-art algorithms, including IC-PCP, ProLiS, PSO, ADBRKGA and L-ACO. Experimental results on five real-world scientific workflows demonstrate that W-LA outperforms the five algorithms on average by 41.33%, 33.29%, 96.88%, 86.37% and 14.36% in terms of WEC.