Cloud dependability analysis: Characterizing Google cluster infrastructure reliability

Cloud computing data centers offer high available and reliable infrastructures for hosting critical applications and data. These data centers host hundreds of thousands physical machines to response to incoming workload as job executing. In this paper, we analyze the Google cloud cluster properties to investigate the relationship among machine failures, updates, and job failures. We present the statistical properties of Google machines and job failures and attempt to correlate them during a 29-day period behave. We classify the machine and job failures per day and represent a reliability model for Google cluster machines using the Continues Time Markov Chains.