To realize automatic analysis and prediction large-scale failure data for high-performance computer systems,an information-theoretic Co-clustering method for large scale nonlinearly correlated failure data was proposed.The failure data were sorted according to the nonlinear correlation of failure features and the failure feature signatures were defined.Then a nonlinearly correlated failure data Co-clustering algorithm was proposed and measured using mutual information entropy.Moreover,the convergence and local optimality of Co-clustering algorithm were proved theoretically.Experimental results on labeled failure data showed that the Co-clustering analysis algorithm outperformed other clustering analysis algorithms and the proposed Co-clustering analysis method had the features of rationality and effectiveness for discovering the nonlinearly correlated failure patterns and could help to predict underlying failures.