Failure analysis and prediction in compound system by wavelets
MetadataShow full item record
The current overall ICT infrastructure mainly the Internet and Telecom networks can be looked upon as an ecosystem, which is the result of the cooperation between a huge number of Autonomous systems (ASes). The interconnection and interdependence between ASes become large and complex as technology advances. This interdependence of ASes or subsystems create vulnerabilities in such a way that problems in one of the interconnected networks affect the normal operation of other networks and even might result in a failure of services across the whole system. The aim of this study is twofold. The first is to discuss about the basic features and trends in the logs of failure data to get some insight about the network s behaviour. In addition to this, the study looks into failure prediction by using the primary failure data to model normal behaviours and predict the system level(critical) failures. Failure log data will be used to model the normal(expected) behaviours of the failures and hence for prediction when there happens a change in the normal behaviour. The report first discusses the conceptual model mainly about some related works as well as a background knowledge on wavelet technique. Then, a simple failure data analysis and brief discussion on the main trend observed during the preliminary study is presented. Lastly, a simple approach for failure prediction using wavelet technique is presented followed by evaluation and discussion of results. The report focuses in using a frequency domain approach which is called wavelet technique. A wavelet based failure prediction approach is proposed which uses some frequencies in the failure log data to characterize the normal operation and hence identify deviations(abnormal behaviours) from the variation in those frequencies when something bad occurs in the network. Once the deviations are identified, a root cause analysis can be conducted for a detail investigation of the problem areas.