Research On Task Scheduling Algorithm In Cloud Environment

With the help of the virtualized computing power, storage resources, and modern Web technology, cloud computing can provide users with extensible, network-centric, abstract IT infrastructure, platform, and a variety of applications. With the depth development in the field of cloud computing application, the system will always produce a large number of tasks and data. How to efficiently schedule tasks and make a distribution has become an important research content in the cloud computing. In this paper, main research works are as follows:(1) In view of the traditional scheduling algorithm, it can’t satisfy the real QoS requirements of users.This paper proposes a task scheduling algorithm based on QoS classification. This algorithm is suitable for independent tasks. First, use fuzzy clustering algorithm to classify task set. Then the traditional segmented Min-Min algorithm is used for task allocation. Segmented Min-Min algorithm is more granular for resource allocation than Min-Min algorithm, so it can improve the task and resource matching degree. Only this way can further reduce the completion time and achieve a certain load balancing. The experimental results show that the proposed method can not only meet the needs of the user’s QoS, but also obtain shorter completion time.(2) Aiming at the basic genetic algorithm fitness computation overhead problem in the case of large population, we propose a fitness evaluation method combined MMTD(medium truth degree). For task allocation, it needs to ensure that all the parent tasks are executed completely, that all parent tasks are executed as quickly as possible and that all the sub tasks are executed in parallel. This can improve the degree of parallelism and reduce the completion time. The main idea of this method is to make an individual fitness evaluation. Firstly, use k-means clustering to cluster the population, then select a representative chromosome in each category and calculate the fitness value. Secondly, select nearest representative chromosome based on representative chromosome. Thirdly, other chromosomes fitness is assessed using MMTD method based on the two categories. Further all chromosomes fitness is concluded. Due to omitting the computing process of the fitness of each chromosome, there will be a corresponding reduction in the computational complexity. The experimental results show that the proposed method can reduce the complexity of computation, and improve the efficiency of iteration.(3) In order to improve data remote access effenciency in MapReduce Model, this method puts forward a kind of optimization scheduling algorithm. In the stage of map, the entire cluster is divided into several parts which at least one data block in each part is copied. According to data interference data interference, tasks are scheduled to data node. The weight of data interference is used to measure resource scarcity. In the reduce stage, delay strategy is adopted to improve the probability of local data access, the optimization measures are taken to minimize the cost of remote data access, which improves the data locality.