Study On Computer Network Fault Diagnose By Data Mining

With the development of computer network, the application of computer network is wider and wider. The reliability and availability of the computer network are be concerned at all the time, especially in military, finance, government and high-tech field. One of the feasible approaches to enhancing the network reliability is to detect and locate the fault in the time because once the failing source is located network administrator can repair the fault as soon as possible. The diagnostic ability of simply network scout is very poor. The application of network fault diagnosis Expert System is widely, but its knowledge is difficult to obtain. Data mining can pick-up the useful knowledge from the mass, incomplete, illegible and random data. Using data mining in the computer network fault diagnose can improve the reliability, stability and agility of network administrate and speed of fault diagnosis. By using data mining, we can implement auto-diagnose of network fault and guarantee high quality of service and reliability.After analyzing the traditional ways of computer network fault diagnose, the thesis proposes a new way of build and update the network fault diagnose expert system's repository. The knowledge was mined from network alarm and performance date.The past researches are focused on the failures of network device; they research the relationship between these failures and locate the fault device. But in computer network's fault, the proportion of network device is on the small side while other fault caused by network virus and network attack is on the large side. So the thesis is focused on the fault beside the network device. We research how to select the network performance parameters, how to collect network alarms and performance parameters, how to select the data mining technique and how to descript the associate rules.By introducing this way into campus network, the author gets a set of rules. After analyzing these associate rules, we found the relationship between network application and performance and the reasons of some network fault. These are difficult to achieve by traditional ways. So the research of this thesis is feasible, correct and practicable.