Research On Task Scheduling Algorithm Based On Load-Balance Aware In Cloud Environment

Cloud computing,as a new high speed network computing service,has been more and more popular.Cloud computing technology is widely used in communication,transportation,finance,manufacturing and other fields.By implementing optimal scheduling of tasks and making full use of available resources to achieve the fastest completion of tasks,it is the goal of the task scheduling algorithm research in cloud computing.With the rapid development of cloud computing,the underlying technical architecture of cloud system has changed significantly,the structure of cloud system is more and more complex,the number of resource nodes is increase,and the difference between the different type clouds is becoming more and more obvious.At the same time,the number of users and industry,the demand for services and timeliness,the data mass and diversity are increasing obviously.The original task scheduling algorithm has been unable to meet the computing requirements,which requires the development of better task scheduling algorithm.In order to meet the new cloud environment and the user need under the new situation,following the basic principle of load balancing,a new adaptive task scheduling algorithm for cloud environment oriented to load balancing is presented.In this thesis,the adaptive task scheduling algorithm based on load balancing in complex cloud environment was systematically studied:The load evaluation method of resource node in private cloud was analyzed and a new evaluation algorithm was presented.The workload prediction problem in the public cloud was analyzed and a new prediction algorithm was presented.Nodes and tasks with different attributes were classified by cluster analysis.According to the results of load forecast and cluster analysis,a set of task scheduling algorithms were presented.The main research results are as follows:(1)A new SARIMA algorithm for node load evaluation specific to the heterogeneity of load caused by the heterogeneity of private cloud nodes was presented.In order to improve the resource utilization ratio of private cloud system,the two-step method based on time series was used to predict CPU workload.WPD method was used to transform the original sequence into more stable subsequences.The accuracy of ARIMA model prediction was improved by SVM fitting,and the load of resource nodes was successfully predicted.In the private cloud environment,the resource node load evaluation results obtained by the SARIMA algorithm could be used as the basis of task scheduling,and the results also could be used to monitor the resource node load.(2)Public cloud usually includes many data centers and resource nodes,and a load forecasting algorithm called GFCEM was presented for the public cloud.The triangular fuzzy weights are used to assign the indexes of each resource node,and the methods of fuzzy comprehensive evaluation and grey correlation theory were used to predict the resource node load in public cloud environment.The GFCEM algorithm could take into account all the factors that may affect the node loads,and improve the accuracy and rationality of the load forecasting results.In public cloud environment,the load results predicted by GFCEM could be used as the initial parameters of task scheduling algorithms,and could also be used as the basis for virtual machine migration.(3)Hybrid Cloud computing are dynamic,high timeliness,and often has the characteristics of numerical and taxonomic multi-attributes,and the number of resource nodes and cloud tasks in cloud computing always are large.The MATC cluster algorithm based on mixed attribute was proposed to optimize task scheduling algorithm,which could carry out cluster analysis of high-dimensional mixed attribute objects.Firstly,the dissimilarity between the objects and the classes was measured,and the centroid of the classes was calculated,and the Euclidean distance between the objects and the centroid was calculated by the improved Euclidean distance model.Then,based on the theory of entropy,the information entropy intra class and inter class were calculated,and the clustering results of mixed attribute objects based on information entropy were obtained.The MATC algorithm could accurately cluster the mixed attribute objects,and the results can be used as the basic initial parameters of cloud task scheduling.(4)As the existing task scheduling algorithm can't fully meet the needs of large scale difference of cloud computing system and the outstanding heterogeneity of resource nodes,load balancing adaptive task scheduling algorithms IAACO and IAPSO were proposed,both algorithms were considered well the factors effect resource node load balancing.The IAACO algorithm mainly solved the phenomenon of the local optimal solution in the traditional algorithm under the premise of satisfying the load balancing.Considering the nodes multiple attributes and obvious differences,the IAPSO algorithm completed the task scheduling on the basis of accurately predicting the load of the nodes,and the optimal task completion time was finally obtained.Based on the above research,a set of new task scheduling algorithms in the hybrid cloud environment was proposed to meet the demands in new situation.The experimental results show that the scheme here has obvious advantages of keeping the system load balance,improving the utilization ratio and execution efficiency of the cloud system.The new algorithms could be widely used in cloud computing environment.