The Research On Resource Scheduling Strategy And Algorithm Of Cloud Computing Platform Based On Local Communication Network

Currently, telecom operators face of computing tasks tend to be complex and diverse, they need to take full advantage of the existing powerful hardware platform to construct a unified cloud computing platform to deal with a variety of computing needs. Cloud computing is a hot research topic of computer science in recent years, it is a distributed super-computing model which based on internet, and it is known as the core of the next generation of Internet technology. Hadoop is one of the cloud computing platform. Hadoop distributed cloud computing platform is a typical cloud computing infrastructure platform, due to it has features of flexible, economical, open source and extended, that make the platform be an important option to build cloud platform of Telecom Operators. Therefore, this paper study, analysis and optimize Hadoop platform based on the computing needs of the telecom operators.Firstly, the paper in-depth study of the main resource scheduling node JT single point failures of Hadoop architecture base on the telecom operators large-scale cluster environment, and proposed a double-JobTracker model,which is exoticed and hot backuped, for large-scale Hadoop cluster. Through the main backup JT election algorithm, to implement two master nodes competition; through analysis and study on responses of the primary node in the damaged situation to ensure the stability of the model; through the backup node to pretreat the feedback information of compute nodes, reducing the processing burden of the master node, and further deepen the coordination of primary and backup nodes.Secondly, the paper research and analysis the existing Hadoop job resource scheduling algorithms and the actual case which is full of numerous burst real-time jobs in Telecom operators. On the basis of the existing algorithm, we propose a new seize model of fair scheduling algorithm based on multi-factor judgement. In order to adapt to different application requirements, the module will target to select the preempted computer notes based on data localization, sufficient using of Resources, attention to job efficiency and attention to the system efficiency etc.Finally, this paper has tested and verified the feasibility and effectiveness of the new preemption which is based on multi-factor model by setting up the simulation environment, and also puts forward corresponding solutions and future research directions according to the experiment phenomenon at the meaning of time.