A prediction approach for correlated failures in distributed computing systems

Abstract

Failure instances in distributed computing systems (DCSs) have exhibited temporal and spatial correlations, where a single failure instance can trigger a set of failure instances simultaneously or successively within a short time interval. We investigate an effective approach to predict correlated failures of computing elements (CEs) in DCSs. Correlated-failure patterns are modeled using the concept of probabilistic shared risk groups (PSRG). Firstly, we design a new structure for PSRG, named SPSRG, to describe features of correlated failures. Then we exploit an association rule mining technique in a parallel way to generate and update our SPSRG using information of CE-failure states. Finally, we propose a correlated failure prediction approach to evaluate the probabilities of upcoming failures from the SPSRG. The experimental results show that the proposed approach outperforms other approaches in failure prediction performance in terms of precision, recall and F-measure. Moreover, it allows employing customizable thresholds by which the trade-off between precision and recall can be adjusted for various requirements.