Research On Task Scheduling Algorithms In Cloud Computing Based On Resource Clustering

With the development of computer technology, network technology and virtualization technology, computing resources are becoming more and more abundant. At the same time,with the acceleration of the pace of life and the improvement of living standards, the requirement of people for the convenience and availability of network resources are more and more strict. In order to meet the needs of the people, cloud computing gradually rise and rapidly develop. Cloud computing is on-demand billing computing model. It provides consumers with great convenience, but also brings great benefits to cloud service providers. Cloud computing concentrate the massive data and computation to the cloud and the data storage and computation in the cloud need a lot of resources. If the scheduling algorithm of cloud system is not reasonable, some resources will be wasted and even the system may not run stably. So, improving the scheduling algorithm is important and necessary.Task scheduling in cloud computing is mainly assigning the resources that meet theconditions to tasks based on demands of customer and aims of system. At present,there are a variety of task scheduling algorithms, which have their own advantages for different task scheduling algorithms, but there are still some shortcomings. In many task scheduling algorithms, the characteristics of resources are not included in the analysis and the demand analysis of the task is not detailed enough, which results in low efficiency of resources and resource utilization. Besides, a part of task scheduling algorithms can be improved in reducing the waiting time of tasks. Aiming at the deficiency of the previous, this paper will classify resources by the improved fuzzy clustering algorithm and combine the fuzzy clustering algorithm with the task scheduling algorithm to assign tasks. Specific research results are as follows:(1) To assign independent tasks, we have combined the kernel-based fuzzy clustering algorithm with the improved FIFO algorithm. Resources will be divided into some clusters by using fuzzy clustering algorithm to calculate the similarity between them. We have optimized the kernel function of the fuzzy clustering algorithm and applied that into classifying resources. Tasks will be assigned to different queues by using new methods to analyze the resource expectation of tasks.By comparing the total finish time of the task on different virtual machines (vm), the task will be reassigned to a queue to wait for resources. By applying the proposed algorithm, the execution time of tasks will be reduced and the utilization of resources will be increased.(2) To assign workflow tasks, we have combined the improved fuzzy clustering algorithm with the improved ant colony optimization algorithm. Workflow tasks have precedence constraints between each other and they can be presented by DAG and described as two-tuples. Resources can be classified by their similarity based on IFCM to reduce the impact of noise and outliers. Workflow tasks can and sorted by the priority. After calculating the distance between the resource expectation of tasks and resources, the original solution will be designed and tasks will be encoded. We have improved ant colony optimization algorithm by the crossover and mutation in genetic algorithm and used the improved algorithm to get the approximate optimal solution. Based on the final solution, the execution time of tasks will be reduced and the utilization of resources will be increased.In addition, this paper has used CloudSim and Hadoop to realize the proposed algorithm and compared the proposed algorithm with FIFO and ACO. The result of experiment shows that the proposed algorithm will do better than FIFO and ACO in reducing total finish time of tasks and improving resource utilization. At the same time, the result also shows that classifying resources by fuzzy clustering algorithm can improve the scheduling performance.