Link Performance Inference Algorithm For Non-Stationary Network

As one of the core functions of the network management, performance management includes the monitoring of the network as well as applications to detect network congestion or disruption, comprehensive performance troubleshooting, capacity planning and efficient allocation of network resources. It occupies a very important position in the network management. Performance management needs to get the status of the network. Currently, the network measurement is the main method to get network status. The network tomography which introduces computer tomography in medicine to the network measurement uses end-to-end measurement at the network boundary to inference the internal performance of the network by analyzing the measurement information. Packet loss rate tomography technology infers link packet loss rate by measuring the end-to-end path packet loss rate. However, the packet loss rate measurement needs to send a large number of packets, so it is difficult to adapt to the rapid change of the non-stationary network and packet loss rate for each link cannot be uniquely identified in most cases, even if we have measured packet loss rate of all paths. Thus people need an efficient network tomography method, which not only be capable with non-stationary network, but can also quickly infer link loss rate according.This paper researches on the key issues of packet loss rate tomography technology, provides more efficient probing and inference methods to address these issues, and verifies the correctness and feasibility of the proposed method through experiment. The main contents include:1) Research and summarize the defects of the existing network tomography techniques, and point out the direction of improvement.2) For the poor performance of the existing probe method in the non-stationary network, we proposes a new loss rate probe method which is suitable for non-stationary network. This method uses signal processing techniques, so that it can detect the time-varying path loss rate more accurately in the non-stationary network, with no extra network load.3) We propose a new inference algorithm with the nonlinear programming technique to solve the performance, accuracy and environment problems of the existing loss rate inference algorithm.4) Verify the correctness and feasibility of proposed method by simulation and real network experiments, and compare our method with existing detection and inference methods.