%0 Journal Article
%A HAN Jin-jin
%A LIU Chun-hong
%A SHANG Yan-lei
%T Predicting Job Failure in Cloud Cluster: Based on SVM Classification
%D 2016
%R 10.13190/j.jbupt.2016.05.021
%J Journal of Beijing University of Posts and Telecommunications
%P 104-109
%V 39
%N 5
%X A job failure predicting method based on support vector machine (SVM) model was presented. Google cluster traces were studied. The relevant factors of jobs failure were analyzed and the combination of the static and dynamic characteristic was chosen as the feature vectors. The SVM algorithm was chosen to predict termination status of the jobs. Experiments were conducted to compare different kinds of feature vectors and classification models with Google traces dataset in terms of the accuracy rate, false negative rate and precision rate. It is shown that the combination of static and dynamic features are 0.94%, -0.01% and 1.35% higher than the static features, and 9.08%, -1.36% and 8.91% higher than the dynamic features. Experiments also demonstrate that the SVM model is superior to the traditional neural network extreme machine learning, naive Bayes and logistic regression model in these indexes.
%U https://journal.bupt.edu.cn/EN/10.13190/j.jbupt.2016.05.021