Research On Bionic Autonomic Monitoring System And Scheduling Mechanism For Balancing Multiple Criterion In Cloud Computing Environment

With the rapid development of cloud computing,the distributed IT resources tend to be centralized.The scale and complexity of the Cloud Computing Environments(CCE)are rising rapidly,which brought many new challenges to cloud resource management.Some of the most important issues are as follows: how to improve the reliability of CCE;how to evaluate the performance of CCE;how to deal with the high energy consumption of CCE;and how to optimize the resource allocation according to user requirement.Towards these issues,they still needs further studies on reliable computing,high performance computing and green computing.It is necessary to propose a correlation model,which unifies reliability,performance and energy.The multidimensional correlation model enables a new scheduling algorithm.In order to improve reliability,monitoring is a necessary means.But,current monitoring system lack self-management mechanism and its structure is not fit for CCE.For system performance evaluation,we must fully consider the characteristics of cloud services in order to establish a suitable performance model,and effectively consider the impact of reliability on performance in order to more accurately assess the performance of cloud services.For green computing,the traditional method only consider energy consumption and performance as a factor of mutual restraint,while ignoring the reliability.In the existing studies,the evaluation of the separation of reliability,performance and energy dissipation is also led to the limitations of the optimal scheduling based on the relevant model.In order to deal with these problems,this thesis studied the monitoring system and scheduling mechanism based on multidimensional correlation model in CCE.The main studies and contributions are as follows:1)For solving the problem that the traditional monitoring system is poorly scalable,lacks autonomy,and can not fully consider the characteristic of cloud computing virtualization and dynamic flexibility,this thesis designed a cloud monitoring system based on Bionic Autonomic Nervous System.The proposed monitor consists of five layers,i.e.,the human-computer interaction layer,the central nervous system layer,the peripheral nervous system layer,the neuronal layer,and the neural ganglion layer.The proposed monitor has the abilities of self-organization,self-diagnosis,self-repair and self-action.The structure of peripheral nerves and neurons fully reflects the characteristics of virtualized resources.The autonomy of the surrounding nervous system can fully reduce the central pressure.In addition,the extendable tree structure of the proposed monitor enables high scalability.Through the self-organization function to effectively solve the difficulty of monitoring and cumbersome problems of deployment due to dynamic changes(join,exit,etc.)of the scale of virtual resources.2)For solving the problem that the traditional performance evaluation ignores the reliability,the thesis proposed a performability model correlated the ”reliability-performance” oriented cloud service feature.The cloud computing environment has heterogeneous,virtualized and on-demand service characteristics,so its relevant performance model(Performability)is different.Considering the failure and repair behavior of virtual resources,using Markov chain to evaluate the reliability,the method of using Universal Generating Function(UGF)to analyze the reliability of heterogeneous cloud resource pool is put forward.Considering the requirement of on-demand service,analyze the different service performance in 1:1 and 1:D mapping mode,the performance model of cloud service system which reflects the process of birth and death is established.Considering the impact of resource reliability randomness on service performance randomness,through Bayesian method to correlate reliability model with performance model,establish a new model of performability for cloud computing.Experimental results show that the two-dimensional associated modeling method can more accurately evaluate the service performance of real cloud computing environment.Its significance is also to provide a common method of analyzing different degrees of correlation about performability.3)Aiming at the shortcomings of traditional evaluation methods to separate reliability,performance and energy consumption indicators,as well as the real problem of severe energy consumption faced by cloud computing environment,the thesis proposed a multidimensional correlation models correlated of reliability,performance and energy and set comprehensive evaluation indicators.The idea of hierarchical modeling is that proposing and resolving the global model of a complex system is often more complex.The method of the thesis is that decompose the global model into multiple sub-models covering the interaction factor,then the global model can be obtained by solving the iterative solution of the sub-models,and complete the overall assessment of the system.In the cloud computing environment,there is a common conditional parameter-”the random number of available resources”,and the sub-models also established is based on the conditional parameters.Finally,the Bayesian theory can be used to remove the conditional parameters,through integrating each sub-models,may obtain a comprehensive evaluation index of multiple indicators associated with expected performance and expected energy consumption.In this paper,the energy consumption model obtained by statistical analysis based on real sampling data is correlated with reliability and performance,and the effectiveness of multidimensional energy consumption assessment is verified by experiments.4)A hierarchical optimization scheduling model and a multi-index joint optimization scheduling strategy are proposed for the complex scheduling requirements of supply and demand constraint faced by the cloud computing environment,the lack of autonomy of the existing hierarchical scheduling and of diversity of optimization scheduling objectives.The specific working mechanism are as follows.The global agent is responsible for the global task distribution,which embodies the centralized control;the local agent realizes the regional autonomy,through the local autonomic computing to achieve”supply and demand map” of local resources optimization reflecting different allocation of the tasks.Through the multi-index joint profit target optimization,To achieve a reasonable allocation of resources to match the global agent dynamic changes in the task requirements.On the basis of local optimization,we use the optimization function which reflects the multidimensional correlation models,and design the concrete genetic algorithm to solve the problem,realize the global optimal under the multi-objective joint optimization.The experimental results show that the hierarchical optimal scheduling mechanism can improve the efficiency of searching the optimal solution and obtain the integrated optimization effect.