Service-oriented Network Fault Diagnosis Based On Fuzzy Association Rule Mining

In the modern information age, even a minor fault in communication networks willbring strong affects on social and economic activities of human beings. Therefore, theintelligent network fault diagnosis is a vital guarantee of next-generation networks.Once alarms occur, it is important for the network operator to identify the alarm typesand levels a.s.a.p, and accurately locate the root alarms to recover networkcommunications promptly. Despite the fact that traditional networks take networkdevices as managed objects, users care more about the normality of network services.However, the normal operation of devices does not reflect the normal operation ofcorresponding services, which means fault managements of services lack user-orientedfeatures. On the other hand, network management experiences two stages as:fundamental platform construction and system integration. It makes device-based faultmanagement no longer suitable for the developing requirements of the communicationnetwork, and there is an inevitable trend of service-oriented network management.This project takes network services as main managed objects, and considersnetwork alarms that contain service information as carriers to study the mappingrelationship between service-oriented QoS parameters and device-oriented networkperformance parameters to get the association of device alarms and service alarmsaccording to the above relationship. Meanwhile, fuzzy theory, fuzzy reasoning and datamining are combined together to realize dynamic fuzzy association rule mining ofmulti-layer multi-domain network fault information.The author discusses multi-layer fuzzy association rule mining, distributed fuzzyassociation rule mining and dynamic fuzzy association rule mining, comprehensivelystudies network fault correlation analysis, at last develops fuzzy association rule basedfuzzy reasoning system. The system can deal with situations when alarm information isuncertain, quickly carry out network fault diagnosis, and accurately locate as well asrecover faults to improve the efficiency and performance of communication networks.The innovation of this study is mainly reflected in the following aspects:(1) Apply fuzzy logic and fuzzy membership function to pretreatment and fuzzification of network alarms, which effectively reduce granularity and complexity ofmulti-dimensional network alarms and improve the mining efficiency. Also the actualmeaning of fuzzy membership degree is the relationship between current alarm and rootalarm, which reflects relative importance and influence areas of network alarms.Performing alarm correlation analysis, fuzzy reasoning and network fault diagnosisbased on fuzzy alarms will be more reasonable.(2) According to hierarchical features of network management, integrate layerinformation when generating alarms. Once the services are affected by network faults,QoS faults of applications are mapped into associated performance faults of networks.(3) Propose the idea of parallel mining distributed fuzzy association rules.Differing from traditional distributed mining algorithms, which horizontally partitionhuge database and use multiple processors for parallel mining rules in sub-database, thisis a trade off between resource and efficiency. This study vertically partition database,and collect alarm information according to time dimension in both global site andlocal site, where the former is in charge of mining inter-network association rules whilethe latter is in charge of mining intra-network association rules. Then develop multi-level multi-dimensional distributed fuzzy association rule mining algorithm–MMDFARM, which considers different sides of algorithm adaptability for networkmanagement in complex environment.(4) In practical environments, the device information, the topology of network andthe service requirements change dynamically, which result in changes of the alarmdatabase. When there is a new alarm, one of the solutions is to remine the wholedatabase to find association rules with the new alarm information. This not only causesa lot of waste of resources, but also completely ignores the knowledge of historicalalarms. The author proposes incremental dynamic fuzzy association rules miningalgorithm–IDFARM according to the time correlation of network alarms, whichconsiders time dimension of algorithm adaptability for the updating communicationnetwork.(5) Deeply study fuzzy association rules based fuzzy reasoning strategy and definemathematic models of fuzzy matching operators, fuzzy implication operators as well asfuzzy composition operators in the process of inference. Experiments are carried out tovalidate the accuracy and efficiency of different combinations of fuzzy operators, and the best one of our synthetic communication network is selected to develop acharacteristic system for network fault diagnosis, which build optimal inference engineby verifying and evaluating various fuzzy reasoning methods and designingfine human-machine interface of network fault management system. Applying fuzzyreasoning to realize automatic and intelligent network management can accuratelydiagnose and well position root alarms, resulting in shortening the recovery time andimproving the performance of communication networks.All in all, realizing quickly and accurately network fault diagnosis based on fuzzyassociation rules of service-oriented alarms and improving intelligent network faultmanagement, are the highlights of this project. In view of the encouraging results, it isthe authors' belief that a fuzzy reasoning will be a valuable tool to assist network faulttroubleshooters in handling the task of diagnosing failure components and for furtherpractical implementation.