It's a hot issue to improve the accuracy of network coordinate systems(NCS). By analyzing the error source of NCSes, it's emphasized that guaranteeing the low dimensionality of embedding datasets is the princi...It's a hot issue to improve the accuracy of network coordinate systems(NCS). By analyzing the error source of NCSes, it's emphasized that guaranteeing the low dimensionality of embedding datasets is the principle on the designing of NCSes. Based on this observation, a PCA based Internet delay space dividing algorithm(IDSD-PCA) is proposed. Beginning at choosing an optimized sub dataset randomly, IDSD-PCA partitions the Internet delay dataset into several sub datasets with low dimensionality by iterations. After the partition, the whole delay dataset and sub datasets are embedded into a hierarchical NCS. The experimental results show that with this framework both the metric space based models and the non-metric space based models can predict the end-to-end delays more accurately.展开更多
基金supported by the National Key Research and Development Program of China under Grant No. 2017YFB0801703the National Natural Science Foundation of China under Grant No. 61602114+1 种基金CERNET Innovation Project No. NGII20170406Jiangsu Provincial Key Laboratory of Computer Network Technology
文摘It's a hot issue to improve the accuracy of network coordinate systems(NCS). By analyzing the error source of NCSes, it's emphasized that guaranteeing the low dimensionality of embedding datasets is the principle on the designing of NCSes. Based on this observation, a PCA based Internet delay space dividing algorithm(IDSD-PCA) is proposed. Beginning at choosing an optimized sub dataset randomly, IDSD-PCA partitions the Internet delay dataset into several sub datasets with low dimensionality by iterations. After the partition, the whole delay dataset and sub datasets are embedded into a hierarchical NCS. The experimental results show that with this framework both the metric space based models and the non-metric space based models can predict the end-to-end delays more accurately.
