A Dimension Separation Based Hybrid Classifier Ensemble for Locating Faults in Cloud Services

Abstract

<p>Cloud services provide Internet users with various services featured with data fusion through the dynamic and expandable virtual resources. Because a large amount of data runs in different modules of the cloud service systems, it will inevitably produce all kinds of failures when the data is processed in and transferred between modules. Therefore the job of rapid fault location has an important role in improving the quality of cloud services. Because of the features of large scale and data fusion of data in the cloud service system, it is difficult to use the conventional fault locating method to locate the faults quickly. Taking the requirements on the speed of locating faults into account, we will make a clear division to all possible failure causes according to the business phases, and quickly locate the faults by implementing a cascading structure of the neural network ensemble. At last, we conducted an experiment of locating faults in a cloud service system runned by a telecom operator, comparing the proposed hybird classifier ensemble with neural networks trained by separated data subsets and a conventional neural network ensemble based on bagging algorithm. The experiment proved that the neural network ensemble based on dimension separation is effective for locating faults in cloud services.</p>