Research On Fault Management Mechanism For Network Function Virtualization

With the increasing demand for network services,network functional virtualization has emerged.Network virtualization functions decouples the software and hardware resources,so that network managers can establish a service chain of network connection equipment without investment and installation expensive proprietary hardware.Besides,network managers can spend less time to manage the data center,so as to reduce capital cost and operating expenses.However,network function virtualization also brings some new challenges to network management while improving network efficiency.The study of network reliability issues has become an important guarantee for the commercialization of network function virtualization.After virtualizing network functions,virtual nodes are more complex than traditional physical nodes,and the virtualization process also consumes additional system resources.These changes make the network more prone to failure,fault propagation and mutual interference more prone to occur.In this context,thesis studies the two parts of the fault detection and alarm correlation analysis under the network function virtualization environment.For fault detection,because of the difficulty of acquiring tagged data and high dimensionality of data in the virtual environment of network functions,traditional numerical detection techniques and supervised machine learning methods are not suitable for fault detection under the virtualized environment of network functions.Therefore,thesis studies a network fault detection mechanism based on an unsupervised machine learning method.Because the existing detection mechanism has its detection blind spots,thesis proposes an integrated fault detection algorithm.This algorithm integrates the widely used self-organizing map algorithm,local outlier algorithm and isolated forest algorithm.This mechanism can detect the blind spots that single algorithm exists,which effectively improves detection performance and improves the detection accuracy of the algorithm.In addition,thesis also corrected the misdetection caused by changes in the network load,and further improved the detection accuracy.For alarm correlation,due to the layered architecture of NFV,a fault that occurs at the lower level may trigger a series of chained reactions at the upper layer,resulting in the simultaneous outburst of a large number of alarms.At the same time,due to the dynamic changes in NFV environment services,it is difficult to obtain the mapping relationship between network elements,so the underlying network topology information is not available.According to the characteristics of alarm correlation analysis in network function virtualization environment,thesis analyzes and compares existing association rules algorithm,and proposes Improved-WINEPI algorithm using FP-growth algorithm in order to accelerate the binomial frequent set searches.Thesis also uses the time correlation of father and son alarm to verify the rules of the Improved-WINEPI algorithm to remove invalid rules and improve the accuracy of association rules.After obtaining the association rules,this thesis uses the root cause diagram for root cause analysis to find root cause alarm and help network administrator to find fault source quickly.