The Research Of Predicting-based Cluster Resource Management

The prediction technology is widely used in every field in our life, such as the in weather field, the business field, and the geography field. The use and development of prediction technology make it possible for people to master the trend of the things in the future time. And so people can arrange their life and work through the forecast information.In the computer science field, the prediction technology is also widely used and more and more research is made on it. The current varies of resource predict technologies have their own merit and demerit which are fit for different type of conditions. Theoretically, it is impossible to design such an arithmetic that can accurately predict all the running circumstance. A new arithmetic of resource predicting technology which we call self-adapting resource predicting arithmetic takes all the popular arithmetic together shows thinking like that: we may congregate all the current arithmetic together, adopt some kind of selection policy, and use the different arithmetic to predict the resource utilization under different circumstances. This arithmetic is aimed to put lot of arithmetic together so that it may have all the current arithmetic's advantage. So it avoids using one kind of arithmetic to predict all of the running circumstances.The job and resource management system is the most important part in the cluster. It aims to integrate all the dispersive resources in all the nodes and provide a single system image to the users. It is highly related to the performance of the cluster.The current job and resource management systems normally adopt the traditional C/S (Client/Server) model, and pay their attention to the current resource information when making their scheduling and load balancing policies. In doing this, it is obviously faulted because the jobs scheduled by the system are running in the "future" and the system's load balance is the "future's" system load balance. The new kind of job & resource management systems must be designed to be more convenient and simple for the users to use, and supply more system information for the administrator to manage. In order to do this, the system may adopt B/S (Browser/Server) model and use the prediction technology we talked about above. After a long time of predicting performance testing, the performance of the self-adapting predicting arithmetic is analyzed and compared with the widely used arithmetic nowadays. In such a system, furthermore, the resource prediction information can be used to make the policy of job scheduling and resource management.