Intelligent Network Performance Management System

The network performance management (NPM) with responsibility for maintaining Quality Of Service and running efficiency is one of basic contents for the network management(NM),and its validity is the all-important problem of NM.First of all, the paper discusses NPM with automatization and distribution and intelligence essential to the current complex interconnected networks.Based on the WEB/CORBA/SNMP teconologies, the paper puts forward a kind of four layed network management model (FLNMM), which is compared with traditional models, and analyses FLNMM's probability and superiority in the complex and isomerous networks.Then, one concentrative and distributed architecture(CDA) based on the four layered model is designed for NPM.In the CDA, domain managers lighten the demand for the center manager's power and less network delay, and at the same time, the center manager can attemper optimal resources because of its holding details of networks.The paper introduces Data Mining and uses Class Mining to build one intelligent decision-machine(IDM) which can mine valuable, hided and impersonal laws in form of patterns and rules from much data which can not be used at first hand by the current analytical tools. Analytical tools can do correct decisions by these rules and patterns advoiding some manjnade destructive results.Based on FLNMM and IDM, one intelligent network performance management system on CORBA(CINPMS) is designed and realized. The system runs as follows: it gathers data from network settings and carries the data to record center. After sorted, these data is showed on controling interface in the form of charts by which users can easily know the running status of networks. If some data reaches or exceeds warning valves, alarms in form of emails and sounds are sended to inform correlative users and at the same time intelligent analyzers make use of rules and patterns mined byIDM to in good time diagnose performance problems . In addition analyzers periodically predict performance trends. In CINPMS , performance problems are solved corresponding to the way in which unsolved problems in the low-lyingdomains are handed over to the high-level performace analyzer. In evidence, CINPMS is characteristic with automatization and distribution and intelligence.