Research On Network Performance Anomaly Detection Method Based On Round Trip Time

With the rapid growth of Internet technology and network service, the computer network is more and more popular and is widely used. The computer network has effect on people's life and working, and the consumers demand the higher performance of network service. The performance of network attracts consumers'attention. The significance of network management is increased recently, one key role of which is performance anomaly detection technology. This technology can detect the problem of network performance and it can ensure usability and high performance of network service.This thesis uses the round trip time (RTT) as performance parameter to investigate the performance anomaly detection technology on IP end-to-end network. The main contents are:Firstly, this thesis puts forward a detection method for network performance anomaly based on trend analysis. We use the least square method and the slipping data window to analyze the historical RTT data based on the character of the RTT distribution, and obtain the normal trend model of RTT. In the process of the network monitoring, we decompound the real-time RTT data into normal trend model component and random component to decrease the influence of periodically normal trend model data on anomaly detection. Using the method of averaging the slipping windows for random component to perform anomaly detection, it makes detection method sensitivity to the anomaly data and the efficiency of anomaly detection is higher.Secondly, this thesis analyzes the deficiency of the periodically sampling. Though the method of periodically sampling applied in the RTT collection is simple, it neglects the data change and collection data can not show the character of RTT data irregular changing area completely. So this thesis puts forward a Poisson sampling method based on link stabilization, gets the expectation sampling interval of Poisson sampling from the parameter of link stabilization, and then calculates actual sampling interval. Using the Poisson sampling, the sampling process is more random, and it can show RTT variation character well. By using sampled data array variance and expectation to estimate the abnormity, we can find the performance problem as soon as possible. The network administrators can take action quickly to prevent the...