| With the rapid development of information technology, the internal structure of processor changes. In recent years the multi-core processor has been popularized gradually, research on multi-core system also become deeper. As the cores in multi-core processor sharing resources, application deployment in different cores will influences each other, cause a decline in each other’s performance. Combined with the data center resource management needs, how to deploy application more rationally, saving energy at the same time make sure the performance of the application meet its QoS, has become the research focus. Due to the elimination of cross-core interference is very difficult and not easy to realize, most of the current studies are focused on the prediction of cross-core interference by the performance of application.Based on the above background, this thesis puts forward application deployment method based on cross-core performance interference on multi-core processor. This method, add the application performance prediction process in the process of application deployment. First of all, analysis the cross-core share resource usage and the performance degradation of application, by the method of correlation analysis and the principal component analysis method, select the cross-core measurement pressure indexes; then, using stepwise regression analysis method to establish model of the application performance degradation and pressure indexes; Finally, by modifying the server ideal pressure value, evaluate the server’s pressure when there are several applications assessment in the server, so as to predict the application performance degradation degree. This thesis also added to predict the performance of the original applications on the server, improve the deployment process. In addition, in order to reduce the prediction cost, when we build the application model, we get the application pressure indexes through sampling method. In the training phase, using the K-means clustering algorithm, select the most suitable data point as training points for modeling, reduce the modeling time.The experimental results show that the accuracy of application deployment method based on cross-core performance interference on multi-core processor. In this thesis, the selected pressure indexes can be competitive useful to measure application’s pressure to the cross-core sharing resources, and the performance model established in the thesis also can accurately predict the performance of application under different pressure. |
