Design And Implementation Of Network Reliability Analysis System Based On Network Traffic Acquisition And Parameters Measurement

With the rapid development of technology of computer and communication, the computer network has penetrated to every aspects of our life. Due to the increasing complication of network structure, reliability becomes an important factor that must be considered in the design and analysis of the network.In this thesis, after analyzing the actuality of the measurement technology and performance requirement of the network, a network reliability analysis system based on network traffic acquisition and parameters measurement is designed and implemented. Key implementation technologies are introduced at the same time. The system has the distributed measurement and concentric analysis architecture. Based on the modules of network topology exploration, network traffic acquisition, network performance parameters measurement and network reliability calculation, the network reliability can be analyzed by the system.Firstly, the network topology discovery module including connection relations and active mainframes is realized. The network routing equipment and subnet are discovered with SNMP protocol. The drawbacks of discovering the active mainframes through the ICMP protocol and the ARP table are pointed out. Also, shortcomings when mainframes have masked the ICMP packets and the real-time flaws with ARP table are described. Thereby, the measurement of active mainframes is realized by sending out UDP data packets and examining the back messages of ICMP protocol. Then the network topology graph is generated.Secondly, self-adaptive network traffic acquisition, statistics of network traffic and real-time alert are implemented in the system. Because the methods of current network traffic acquisition have the defects of discovering the network status inaccurately and increasing the network traffic, a self-adaptive algorithm is chosen, which can adjust the acquisition interval according to the range of traffic data variation. This algorithm is implemented by the asynchronous sockets in C#. The network traffic data in the database are used for statistics such as the statistic of protocol and TOP. Based on the simple method of clustering algorithm, the traffic distributing is found out. At the same time, under the 3σanomaly detection principle, the anticipation of real-time network traffic is inspected.Finally, performance parameter measurement is accomplished by sending packets actively and the results of network reliability are analyzed. For the requirement of experimental network reliability, the packet delay, delay variation and loss probability of data packets are measured by sending the modified data packets of UDP protocol. The shortage of traditional parameters evaluation methods which include the fixed threshold method and the dynamic threshold method (by means of scheduling and history records respectively) is studied under experiments. Then a self-adaptive dynamic threshold statistical method with mixed scheduling and history records is proposed in this thesis. It can provide the threshold accurately. After indicating the shortcoming of assigning weights averagely for multi-parameters, and normalizing the parameters, a new method which assigns different weights for multi-parameters according to applications is introduced.The measurement experiments under actual network show that the system is efficient for network reliability measurement.