Research On Resource Provisioning Mechanisms For Cloud Services Based On Service Selection

The emergence of cloud computing makes it possible for service providers providing on-demand services.Applications in cloud environment are presented to users as cloud services,and for cloud service providers,they usually want to obtain commercial profits by cutting down resources' using cost while meeting performance requirements of users,then the problem of resource provision for cloud services is generated.In order to solve this issue,many researches according to different measures have done.But they usually take no account of the structure of the cloud services,in addition,what they concern are dynamic environments.There is lack of work going into optimal resource allocation for cloud service decribed as service based software system(SBS)in many combinations.This thesis takes SBS as a typical representative of cloud services for research.To solve the problem of SBS initial resource provision,this thesis proposes a method based on service selection.This method translates SBS resource provision problem into service selection problem.Each component service has a set of optional logical services and optimal resource provision among optional logical services is determined for every component service by service selection mothod.This thesis is divided into three parts:discretization of resource attributes?modeling of resource-performance relationship?resource provision optimization model and solving algorithm.In discreting resource attributes,this thesis studies the method of discreting four kinds of resource attributes;analyzes the relationship between service performance and various resource attributes.Based on this,the CAIM algotithm is used to generate collection of discreted resource attributes,so that it can essentially support further research.In modeling of resource-performance relationship,this thesis analyzes the characteristics of initial resource provision,collaborative filitering method is used to deal with the sparse data in benchmark test.Then BP neural network is used to build relational model between resource attributes and services' performance,this model is used to predict the performance of component service when a certain number of resources are allocated and thus optional logical services set is generated.Based on optional logical services of component service,the resource provision for SBS is translated into optimal service selection problem.Genetic algorithm is used to solve the optimization problem.Eventually,a complete SBS instance is used and some experiments are designed to prove the practicality and effectiveness of the method for service selection based resource provision of SBS.