| IVCE(Internet-based Virtual Computing Environment)platform is an Internet based virtual computing environment,which mainly provides users with efficient and reliable services,and enables them to make full use of and share public resources on the platform.As the platform system is complex and the task concurrency is high,how to schedule tasks reasonably and provide high quality cloud services is one of the key technologies of IVCE platform.In addition,the platform has many tasks,vague needs,and various kinds of categories,how to excavate the characteristics of tasks and find links between tasks has become an important research direction of IVCE platform.Aiming at the above problems existing in the IVCE platform,this paper first adopts the improved clustering algorithm to cluster tasks to find commonality of them;then,the reinforcement learning algorithm is applied to the task scheduling to improve overall operating efficiency and quality of service platform;finally,a task scheduling system based on above theories is designed and implemented.The main contents of this paper are as follows:(1)An improved kernel k-means algorithm is proposed and applied to task clustering.Traditional kernel kmeans algorithm has two problems:First,the k value needs to be artificially determined and can not be changed during clustering.When faced with high-dimensional,massive data sets,it is difficult to accurately estimate the size of the k value.Second,for the sample data that needs to be clustered,the number of each type of sample is different,and the importance of each sample point is also different.Therefore,clustering results are vulnerable to the majority of sample points,reducing the accuracy of clustering results.In view of the above problems,this paper proposes an improved kernel kmeans algorithm,that is Adaptive Kernel Kmeans.The algorithm can adjust the value of k in a certain range according to the actual situation,including splitting and merging two operation,while weighting the distance function to distinguish the importance of different samples.According to the experimental results of UCI dataset,the accuracy of clustering can be improved by 11%and the number of iterations reduced by 16.7%compared with kernel k-means algorithm.Finally,the algorithm divides the 9 tasks of IVCE platform into 5 categories and tests the execution time before and after task clustering.The average execution time of tasks is reduced by 23%,indicating that the improved algorithm can better mine the tasks Commonalities between.(2)Proposed task scheduling strategy based on enhanced learning algorithm.IVCE platform to solve the task and funding the problem of low efficiency caused by the low degree of source matching,this paper proposes to apply QLearning in the enhanced learning algorithm to the task scheduling strategy.Through the establishment of task scheduling model,the definition of system state,behavior and return function to solve the problem of resource matching.In order to speed up the solution,this paper uses BP neural network for nonlinear approximation,reducing the scheduling algorithm time.Experimental results show that the scheduling strategy based on QLearning algorithm outperforms other scheduling algorithms in terms of task dwell time,task success rate and resource consumption.Therefore,this strategy can improve the overall efficiency of IVCE cloud platform.(3)Design and implement a task scheduling system based on QLearning.This article finally achieved based on the above theory of task scheduling system.First of all,the overall architecture of the system is designed.Then,the design and implementation of each module in the system are described in detail,including application registration module,task receiving module,task clustering module,task scheduling module and task recovery module.Finally,the basic functions of each module test to verify the correctness and integrity of the system.The test results show that the scheduling system has perfect functions and can realize the reasonable scheduling of tasks. |
PDF Full Text Request