Improving the execution efficiency as well as reducing the execution cost of the scientific workflows in cloud is important. Focused on the conflict between userdesired local QoS constraints and the overall execution efficiency of the workflow，we propose a scheduling policy of scientific workflows allowing the violation of local time constraints. Based on backward merging of task clusters, free time spans between workflow task executions can be exploited, and the whole execution time of the workflow can be optimized. Furthermore，in order to make full use of the slack time during task execution and improve the overall efficiency of the workflow, some workflow tasks are permitted to violate the local constraint of the latest finish time. Experimental results show that the policy can bring forward the earliest finish time of the workflow and improve the utilization ratio of processors, and eventually lower the execution cost of the workflow.

 

scientific workflow, task scheduling, cluster aggregation, slack time, local constraint