Using unsupervised machine learning for fault identification in virtual machines

Machine learning;Virtual computer systems--Quality control;Fault location (Engineering);Neural networks (Computer science);Self organizing systems

Schneider, Christopher ; Dobson, Simon ; Dobson, Simon

University:University of St Andrews

Department:Computer Science (School of)

关键词: Machine learning;    Virtual computer systems--Quality control;    Fault location (Engineering);    Neural networks (Computer science);    Self organizing systems;   

Others  :  https://research-repository.st-andrews.ac.uk/bitstream/handle/10023/7327/ChrisSchneiderPhDThesis.pdf?sequence=8&isAllowed=y

来源: DR-NTU

【 摘 要 】

Self-healing systems promise operating cost reductions in large-scale computingenvironments through the automated detection of, and recovery from, faults.However, at present there appears to be little known empirical evidence comparing thedifferent approaches, or demonstrations that such implementations reduce costs.This thesis compares previous and current self-healing approaches before demonstratinga new, unsupervised approach that combines artificial neural networks withperformance tests to perform fault identification in an automated fashion, i.e. thecorrect and accurate determination of which computer features are associated witha given performance test failure.Several key contributions are made in the course of this research including ananalysis of the different types of self-healing approaches based on their contextualuse, a baseline for future comparisons between self-healing frameworks thatuse artificial neural networks, and a successful, automated fault identification incloud infrastructure, and more specifically virtual machines. This approach usesthree established machine learning techniques: Naïve Bayes, Baum-Welch, andContrastive Divergence Learning. The latter demonstrates minimisation of human-interactionbeyond previous implementations by producing a list in decreasingorder of likelihood of potential root causes (i.e. fault hypotheses) which bringsthe state of the art one step closer toward fully self-healing systems.This thesis also examines the impact of that different types of faults have on theirrespective identification. This helps to understand the validity of the data beingpresented, and how the field is progressing, whilst examining the differences inimpact to identification between emulated thread crashes and errant user changes –a contribution believed to be unique to this research.Lastly, future research avenues and conclusions in automated fault identificationare described along with lessons learned throughout this endeavor. This includesthe progression of artificial neural networks, how learning algorithms are beingdeveloped and understood, and possibilities for automatically generating featurelocality data.