A Failure Prediction Approach Based on Cloud Theory and Hidden Markov Model in Networked Computing Systems

摘要

Due to off-the-shelf hardware and software applications integrated with distinct manufactures are widely used, networked computing systems incur high risk of failures and exceptions. Failures play a crucial role and must be timely handled to ensure system survivability and reliability. This paper focuses on on-line failure prediction for networked computing systems using system runtime data. We propose a failure prediction approach based on cloud theory (CT) and hidden Markov model (HMM). This approach expands the HMM, training with the CT. Additionally, we define the parameter w as the correlations between various indices and failures, taking account of multiple runtime indices in networked computing systems. And we use multiple dimensions to describe failure prediction in detail, by extending parameters in HMM. In order to reduce computing cost in model training phase, we exploit the likelihood and membership degree computing algorithms in CT, instead of traditional HMM algorithms. Finally, the results from our simulations show the feasibilities and effectiveness of our approach. The experiments show that the execution time of the proposed failure prediction is reduced in terms of promised prediction performance.