Deadline Decomposition And Scheduling Of Batch Science Workflow For Cost Optimzation

With the deepening of cloud computing applications and the increasing demand for big data processing,more and more enterprises choose to use cloud platform to process massive data.Due to the commercial nature of cloud computing,it puts forward more stringent requirements for task scheduling in cloud computing.How to reasonably and economically complete task scheduling has become one of the key issues in cloud computing research.Batch scientific workflow is a new form of workflow modeling in the era of big data.The traditional scheduling algorithm of scientific workflow is not suitable for batch scientific workflow.Therefore,in this paper,on the basis of predecessors' research,around a batch of scientific workflow task scheduling model with the Deadline,carries on the discussion and research of the overall scheduling costs,this paper proposes a batch of scientific workflow dynamic Deadline partition method based on the rules of iteration and a batch of scientific workflow based on improved genetic algorithm of task scheduling algorithm.The traditional Deadline partitioning method is not only unable to adapt to the batch scientific workflow,but also leads to the fact that it is too simple to divide the task scheduling time of each part reasonably in the actual task scheduling process,which leads to the great increase of task scheduling cost.Therefore,this paper proposes a dynamic Deadline partitioning method based on rule iteration,which can divide Deadline of batch scientific workflow better by reasonably reducing parallelism or upgrading virtual machine and simultaneously shrinking the number of virtual machine usage on non-critical path.Based on the traditional genetic algorithm,a Deadline partitioning method based on the improved genetic algorithm is proposed to adapt to the new batch scientific workflow.The main improvement is to optimize the generation process of the initial population,so that the generated initial population is better while maintaining the randomness,and the algorithm generally converges faster.The fitness function is optimized to make the algorithm more suitable for batch scientific workflow.At the same time,the non-critical path virtual machine is introduced to reduce the task scheduling cost.Finally,experiments verify the validity of the two Deadline partitioning methods mentioned above in batch scientific workflow task scheduling.The experimental results show that the two partition methods proposed in this paper can reduce the task scheduling cost under the premise of meeting the deadline as far as possible.The research results of this paper can provide new ideas and references for the research on task scheduling of data-intensive batch processing science workflow under the big data environment.