Resource Optimization Method And System For Cloud Computing Center

Through virtualization technology,cloud computing enables flexible allocation of computing and storage resources,etc.The large resources scale,multitask concurrent execution,and dynamic changes of application resource demands make the cloud computing center's resource allocation face severe challenges.Real-time dynamic resource allocation and adjustment is an effective method,which achieves highly efficient allocation and utilization of cloud computing center resources and guarantees quality of service.This paper focuses on the resource allocation optimization of the cloud computing center,realizing realtime optimal utilization of self-adaptation resource between user demands and environments through highly efficient dynamic adjustment.The main research work is as follows:For the resource allocation adjustment algorithms such as Nelder-Mead and hill-climbing,a local optimal method is adopted,and there are problems that it is difficult to quickly respond to dynamic changes of application load demands due to the number of iterations and slow convergence rate.Through the rapid mapping of cloud computing center resources and application load demands,this paper proposes a resource optimization allocation strategy construction method based on reinforcement learning(RL),which achieves real-time response to user load dynamic changes and ensures that the virtual machine hosting the application load keeps high-productive configuration status;The resource dynamic adjustment is divided into three levels including the autonomous layer,the resource layer,and the virtual machine layer,this paper proposes an adaptive resource dynamic elastic adjustment method based on application demands,which improves the resource utilization rate on the basis of ensuring the service quality of the cloud computing center under the dynamic change of resource demands.Experimental results show that compared with the Nelder-Mead method,the proposed method reduces the delay of resource dynamic allocation by 20%.For the learning process of the RL-based resource optimization allocation strategy construction,each iteration uses a random policy ?-greedy to compromise exploration and exploitation two processes to generate a virtual machine resource allocation strategy.During the exploration process,the random allocation of virtual machine resources may reduce the performances of the application hosted by virtual machine,losing the meaning of exploration and affecting learning efficiency.By extracting core system metrics(such as CPU,memory utilization,etc.)to reduce the resource allocation behaviors which may reduce the performances of the application hosted by virtual machine during the exploration process and thereby accelerate the learning process,a strategy construction method based on system knowledge guided exploration process is proposed,improving the efficiency of learning process of resource allocation strategy construction.For the policy-based adaptive dynamic elasticity resource adjustment method,the large search configuration set leads to slower convergence rate of the strategy construction process and a longer search time of virtual machine online adaptive resource dynamic adjustment,a random Nelder-Mead(RNM)algorithm is proposed to reduce the search range to the optimal configuration set,i.e.,most resource configurations in this configuration set can make the application hosted by the virtual machine meet the SLA(Service Level Agreement),thereby improving the algorithm's convergence rate and online resource adjustment efficiency.Based on the above improvements,an RNM-RL resource dynamic elasticity algorithm is proposed,which takes advantage of the global search capabilities of RNM and the scalability and adaptability of the RL algorithm.And it also further enhances the algorithm's learning effectiveness by using system knowledge guided exploration policy than the random policy ?-greedy.Experimental results show that compared with the Nelder-Mead method,the delay of resource dynamic allocation of the proposed method is reduced by 26%.Compared with the policy-based dynamic elasticity resource adjustment method,the proposed method reduces the resource dynamic allocation delay by 5.5%.Based on the research methods,this paper designs an efficient resource optimal allocation system for the cloud computing center.This system has the functions of monitoring application load,resource optimization allocation strategy construction,and dynamic adjustment of resources,which can realize real-time resource adjustment of cloud computing center and guarantee quality of service.And the effectiveness of the research methods is verified by performance tests.