关键词: 性能测量, 网络断层扫描, 链路消减, 系统域网络

Abstract: System area network is the important component of high performance computer and data center. The current system area network has the characteristics of larger-scale network, more internal links, more complex behavior of network traffic, higher sensitivity of applications on network performance and so on. Those problems cause that the computational complexity of the performance measurement is exponential growth of the traditional network tomo-
graphy. To address this problem, this paper proposes a new method of network tomography called NTRS (network tomography based on reduction strategy). This method proposes a number of pre-treatment principles to optimize the measurement topology and these principles include merging the consistent physical links, making full use of the known performance information of internal links, narrowing the scope of the performance measuring area and covering the key links. According to the results of measurement, this method screens for link collection with better performance to achieve an effective reduction of the measuring links through computing the covariance of performance parameters for internal links in light of end-to-end performance measurement data. And the method largely improves the accuracy of performance measurement and the link diagnosis speed. Finally, this paper validates the effectiveness of NTRS by simulation, the results show that NTRS can decrease the number of the link performance parameters needed to measure and reduce the computational complexity.

Key words: performance measurement, network tomography, link reduction, system area network