Research On Optimal Resource Allocation Of Cloud Platform Via Adaptive Decision-making Clustering

Cloud platform has attracted extensive attention in the Internet industry due to its low cost,abundant service resources,strong capabilities of storage and computing.However,with the continuous increase in the number of users on the cloud platform,cloud task data has the characteristics of large scale and multi-source.In addition,due to the high heterogeneity of physical and virtual resources,the platform lacks real-time analysis method of task data.This makes the deployed resources fail to adapt to the changing trend of tasks in a timely and reasonable manner,which results in a shortage or waste of resources and impacting the service performance on the cloud platform.Therefore,how to rely on effective task data analysis methods to achieve the full utilization and reasonable allocation of resources in the platform is the one of key issues to ensure the economic benefits of providers and users.Aiming at the above problems,this thesis proposes an adaptive decision-making clustering method for optimal resource allocation via clustering analysis method as the key technology,focusing on the resource allocation problem in cloud management platform.The main research contents and results of this thesis are as follows:(1)A method of data overlapping decision-making based on subspace partition is proposed.In the allocation process of users' tasks and resources,task data is large in scale and high in dimension.To solve this problem,the idea of K-nearest neighbor is used to highlight local information of nonlinear data structure,and to reduce the complexity of global information search.Then,the overlapping probability model is used to provide decision-making for overlapping parts of data in subspace,and the data samples are adaptively divided into corresponding categories to improve the clustering accuracy.The experimental results show that the method can achieve more accurate subspace partition results and is suitable for practical application.(2)An ensemble clustering method based on distance decision is proposed.Since the single clustering method is sensitive to the initial clustering centers,it can not adapt to a variety of cloud user tasks,which directly affects the optimal allocation of tasks and resources.In order to improve the accuracy and stability of clustering,the fuzzy c-means algorithm is used as the base clustering input,and multiple fuzzy clustering results are combined to establish a cumulative distance matrix and introduced into the density peak algorithm for clustering ensemble.This clustering ensemble method can effectively solve the unstable clustering problem.(3)An adaptive ensemble method based on uncertain entropy decision-making is proposed.The existing clustering methods for optimal resource allocation can not determine the number of clusters adaptively,and the clustering results of task data are inaccurate.To solve these problems,the stable and high quality members are selected from several base clustering members as the ensemble input.Based on the information entropy criterion,the uncertainty of clusters in clustering members is calculated,and the co-association matrix is established.The matrix is introduced into the density peak algorithm to obtain the final clustering results.The decision clustering model established by this method is more accurate and reliable,and further improves the clustering performance.(4)A method of elastic resource provisioning and optimal scheduling based on task clustering in cloud platform is proposed.Aiming at the problem of poor matching of cloud tasks and resources,the proposed clustering methods are used to analyze cloud task data to provide decisions for system resource allocation and task scheduling.For each cluster,the method of time series is used to predict the amount of tasks arriving at the next moment,which provides decision-making for subsequent resource allocation.Then,a resource provisioning method is designed to save system energy consumption,which dynamically provides resources for the tasks of each cluster to achieve the optimal allocation of resources in the system.